View Javadoc

1   /**
2    * Copyright 2010 The Apache Software Foundation
3    *
4    * Licensed to the Apache Software Foundation (ASF) under one
5    * or more contributor license agreements.  See the NOTICE file
6    * distributed with this work for additional information
7    * regarding copyright ownership.  The ASF licenses this file
8    * to you under the Apache License, Version 2.0 (the
9    * "License"); you may not use this file except in compliance
10   * with the License.  You may obtain a copy of the License at
11   *
12   *     http://www.apache.org/licenses/LICENSE-2.0
13   *
14   * Unless required by applicable law or agreed to in writing, software
15   * distributed under the License is distributed on an "AS IS" BASIS,
16   * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
17   * See the License for the specific language governing permissions and
18   * limitations under the License.
19   */
20  package org.apache.hadoop.hbase.master;
21  
22  import java.io.IOException;
23  import java.lang.reflect.Constructor;
24  import java.lang.reflect.InvocationTargetException;
25  import java.net.InetSocketAddress;
26  import java.net.UnknownHostException;
27  import java.util.ArrayList;
28  import java.util.List;
29  import java.util.Map;
30  import java.util.concurrent.atomic.AtomicReference;
31  
32  import org.apache.commons.logging.Log;
33  import org.apache.commons.logging.LogFactory;
34  import org.apache.hadoop.conf.Configuration;
35  import org.apache.hadoop.hbase.Chore;
36  import org.apache.hadoop.hbase.ClusterStatus;
37  import org.apache.hadoop.hbase.HColumnDescriptor;
38  import org.apache.hadoop.hbase.HConstants;
39  import org.apache.hadoop.hbase.HMsg;
40  import org.apache.hadoop.hbase.HRegionInfo;
41  import org.apache.hadoop.hbase.HServerAddress;
42  import org.apache.hadoop.hbase.HServerInfo;
43  import org.apache.hadoop.hbase.HTableDescriptor;
44  import org.apache.hadoop.hbase.MasterNotRunningException;
45  import org.apache.hadoop.hbase.NotAllMetaRegionsOnlineException;
46  import org.apache.hadoop.hbase.Server;
47  import org.apache.hadoop.hbase.TableExistsException;
48  import org.apache.hadoop.hbase.TableNotDisabledException;
49  import org.apache.hadoop.hbase.TableNotFoundException;
50  import org.apache.hadoop.hbase.UnknownRegionException;
51  import org.apache.hadoop.hbase.catalog.CatalogTracker;
52  import org.apache.hadoop.hbase.catalog.MetaEditor;
53  import org.apache.hadoop.hbase.catalog.MetaReader;
54  import org.apache.hadoop.hbase.client.HConnection;
55  import org.apache.hadoop.hbase.client.HConnectionManager;
56  import org.apache.hadoop.hbase.client.MetaScanner;
57  import org.apache.hadoop.hbase.client.MetaScanner.MetaScannerVisitor;
58  import org.apache.hadoop.hbase.client.Result;
59  import org.apache.hadoop.hbase.executor.ExecutorService;
60  import org.apache.hadoop.hbase.executor.ExecutorService.ExecutorType;
61  import org.apache.hadoop.hbase.ipc.HBaseRPC;
62  import org.apache.hadoop.hbase.ipc.HBaseRPCProtocolVersion;
63  import org.apache.hadoop.hbase.ipc.HBaseServer;
64  import org.apache.hadoop.hbase.ipc.HMasterInterface;
65  import org.apache.hadoop.hbase.ipc.HMasterRegionInterface;
66  import org.apache.hadoop.hbase.master.LoadBalancer.RegionPlan;
67  import org.apache.hadoop.hbase.master.handler.DeleteTableHandler;
68  import org.apache.hadoop.hbase.master.handler.DisableTableHandler;
69  import org.apache.hadoop.hbase.master.handler.EnableTableHandler;
70  import org.apache.hadoop.hbase.master.handler.ModifyTableHandler;
71  import org.apache.hadoop.hbase.master.handler.TableAddFamilyHandler;
72  import org.apache.hadoop.hbase.master.handler.TableDeleteFamilyHandler;
73  import org.apache.hadoop.hbase.master.handler.TableModifyFamilyHandler;
74  import org.apache.hadoop.hbase.master.metrics.MasterMetrics;
75  import org.apache.hadoop.hbase.regionserver.HRegion;
76  import org.apache.hadoop.hbase.replication.regionserver.Replication;
77  import org.apache.hadoop.hbase.security.User;
78  import org.apache.hadoop.hbase.util.Bytes;
79  import org.apache.hadoop.hbase.util.InfoServer;
80  import org.apache.hadoop.hbase.util.Pair;
81  import org.apache.hadoop.hbase.util.Sleeper;
82  import org.apache.hadoop.hbase.util.Strings;
83  import org.apache.hadoop.hbase.util.Threads;
84  import org.apache.hadoop.hbase.util.VersionInfo;
85  import org.apache.hadoop.hbase.zookeeper.ClusterStatusTracker;
86  import org.apache.hadoop.hbase.zookeeper.RegionServerTracker;
87  import org.apache.hadoop.hbase.zookeeper.ZooKeeperWatcher;
88  import org.apache.hadoop.io.MapWritable;
89  import org.apache.hadoop.io.Text;
90  import org.apache.hadoop.net.DNS;
91  import org.apache.zookeeper.KeeperException;
92  import org.apache.zookeeper.Watcher;
93  
94  /**
95   * HMaster is the "master server" for HBase. An HBase cluster has one active
96   * master.  If many masters are started, all compete.  Whichever wins goes on to
97   * run the cluster.  All others park themselves in their constructor until
98   * master or cluster shutdown or until the active master loses its lease in
99   * zookeeper.  Thereafter, all running master jostle to take over master role.
100  *
101  * <p>The Master can be asked shutdown the cluster. See {@link #shutdown()}.  In
102  * this case it will tell all regionservers to go down and then wait on them
103  * all reporting in that they are down.  This master will then shut itself down.
104  *
105  * <p>You can also shutdown just this master.  Call {@link #stopMaster()}.
106  *
107  * @see HMasterInterface
108  * @see HMasterRegionInterface
109  * @see Watcher
110  */
111 public class HMaster extends Thread
112 implements HMasterInterface, HMasterRegionInterface, MasterServices, Server {
113   private static final Log LOG = LogFactory.getLog(HMaster.class.getName());
114 
115   // MASTER is name of the webapp and the attribute name used stuffing this
116   //instance into web context.
117   public static final String MASTER = "master";
118 
119   // The configuration for the Master
120   private final Configuration conf;
121   // server for the web ui
122   private InfoServer infoServer;
123 
124   // Our zk client.
125   private ZooKeeperWatcher zooKeeper;
126   // Manager and zk listener for master election
127   private ActiveMasterManager activeMasterManager;
128   // Region server tracker
129   private RegionServerTracker regionServerTracker;
130 
131   // RPC server for the HMaster
132   private final HBaseServer rpcServer;
133   // Address of the HMaster
134   private final HServerAddress address;
135   // Metrics for the HMaster
136   private final MasterMetrics metrics;
137   // file system manager for the master FS operations
138   private MasterFileSystem fileSystemManager;
139 
140   private HConnection connection;
141 
142   // server manager to deal with region server info
143   private ServerManager serverManager;
144 
145   // manager of assignment nodes in zookeeper
146   AssignmentManager assignmentManager;
147   // manager of catalog regions
148   private CatalogTracker catalogTracker;
149   // Cluster status zk tracker and local setter
150   private ClusterStatusTracker clusterStatusTracker;
151 
152   // This flag is for stopping this Master instance.  Its set when we are
153   // stopping or aborting
154   private volatile boolean stopped = false;
155   // Set on abort -- usually failure of our zk session.
156   private volatile boolean abort = false;
157   // flag set after we become the active master (used for testing)
158   private volatile boolean isActiveMaster = false;
159   // flag set after we complete initialization once active (used for testing)
160   private volatile boolean initialized = false;
161 
162   // Instance of the hbase executor service.
163   ExecutorService executorService;
164 
165   private LoadBalancer balancer;
166   private Thread balancerChore;
167   // If 'true', the balancer is 'on'.  If 'false', the balancer will not run.
168   private volatile boolean balanceSwitch = true;
169 
170   private Thread catalogJanitorChore;
171   private LogCleaner logCleaner;
172 
173   /**
174    * Initializes the HMaster. The steps are as follows:
175    * <p>
176    * <ol>
177    * <li>Initialize HMaster RPC and address
178    * <li>Connect to ZooKeeper.
179    * </ol>
180    * <p>
181    * Remaining steps of initialization occur in {@link #run()} so that they
182    * run in their own thread rather than within the context of the constructor.
183    * @throws InterruptedException
184    */
185   public HMaster(final Configuration conf)
186   throws IOException, KeeperException, InterruptedException {
187     this.conf = conf;
188 
189     /*
190      * Determine address and initialize RPC server (but do not start).
191      * The RPC server ports can be ephemeral. Create a ZKW instance.
192      */
193     HServerAddress a = new HServerAddress(getMyAddress(this.conf));
194     int numHandlers = conf.getInt("hbase.regionserver.handler.count", 10);
195     this.rpcServer = HBaseRPC.getServer(this,
196       new Class<?>[]{HMasterInterface.class, HMasterRegionInterface.class},
197       a.getBindAddress(), a.getPort(),
198       numHandlers,
199       0, // we dont use high priority handlers in master
200       false, conf,
201       0); // this is a DNC w/o high priority handlers
202     this.address = new HServerAddress(rpcServer.getListenerAddress());
203 
204     // initialize server principal (if using secure Hadoop)
205     User.login(conf, "hbase.master.keytab.file",
206         "hbase.master.kerberos.principal", this.address.getHostname());
207 
208     // set the thread name now we have an address
209     setName(MASTER + "-" + this.address);
210 
211     Replication.decorateMasterConfiguration(this.conf);
212 
213     this.rpcServer.startThreads();
214 
215     // Hack! Maps DFSClient => Master for logs.  HDFS made this
216     // config param for task trackers, but we can piggyback off of it.
217     if (this.conf.get("mapred.task.id") == null) {
218       this.conf.set("mapred.task.id", "hb_m_" + this.address.toString() +
219         "_" + System.currentTimeMillis());
220     }
221 
222     this.zooKeeper = new ZooKeeperWatcher(conf, MASTER + ":" +
223         address.getPort(), this);
224 
225     this.metrics = new MasterMetrics(getServerName());
226   }
227 
228   /**
229    * Stall startup if we are designated a backup master; i.e. we want someone
230    * else to become the master before proceeding.
231    * @param c
232    * @param amm
233    * @throws InterruptedException
234    */
235   private static void stallIfBackupMaster(final Configuration c,
236       final ActiveMasterManager amm)
237   throws InterruptedException {
238     // If we're a backup master, stall until a primary to writes his address
239     if (!c.getBoolean(HConstants.MASTER_TYPE_BACKUP,
240       HConstants.DEFAULT_MASTER_TYPE_BACKUP)) {
241       return;
242     }
243     LOG.debug("HMaster started in backup mode.  " +
244       "Stalling until master znode is written.");
245     // This will only be a minute or so while the cluster starts up,
246     // so don't worry about setting watches on the parent znode
247     while (!amm.isActiveMaster()) {
248       LOG.debug("Waiting for master address ZNode to be written " +
249         "(Also watching cluster state node)");
250       Thread.sleep(c.getInt("zookeeper.session.timeout", 180 * 1000));
251     }
252   }
253 
254   /**
255    * Main processing loop for the HMaster.
256    * <ol>
257    * <li>Block until becoming active master
258    * <li>Finish initialization via {@link #finishInitialization()}
259    * <li>Enter loop until we are stopped
260    * <li>Stop services and perform cleanup once stopped
261    * </ol>
262    */
263   @Override
264   public void run() {
265     try {
266       /*
267        * Block on becoming the active master.
268        *
269        * We race with other masters to write our address into ZooKeeper.  If we
270        * succeed, we are the primary/active master and finish initialization.
271        *
272        * If we do not succeed, there is another active master and we should
273        * now wait until it dies to try and become the next active master.  If we
274        * do not succeed on our first attempt, this is no longer a cluster startup.
275        */
276       this.activeMasterManager = new ActiveMasterManager(zooKeeper, address, this);
277       this.zooKeeper.registerListener(activeMasterManager);
278       stallIfBackupMaster(this.conf, this.activeMasterManager);
279       this.activeMasterManager.blockUntilBecomingActiveMaster();
280       // We are either the active master or we were asked to shutdown
281       if (!this.stopped) {
282         finishInitialization();
283         loop();
284       }
285     } catch (Throwable t) {
286       abort("Unhandled exception. Starting shutdown.", t);
287     } finally {
288       stopChores();
289       // Wait for all the remaining region servers to report in IFF we were
290       // running a cluster shutdown AND we were NOT aborting.
291       if (!this.abort && this.serverManager != null &&
292           this.serverManager.isClusterShutdown()) {
293         this.serverManager.letRegionServersShutdown();
294       }
295       stopServiceThreads();
296       // Stop services started for both backup and active masters
297       if (this.activeMasterManager != null) this.activeMasterManager.stop();
298       if (this.catalogTracker != null) this.catalogTracker.stop();
299       if (this.serverManager != null) this.serverManager.stop();
300       if (this.assignmentManager != null) this.assignmentManager.stop();
301       HConnectionManager.deleteConnection(this.conf, true);
302       this.zooKeeper.close();
303     }
304     LOG.info("HMaster main thread exiting");
305   }
306 
307   private void loop() {
308     // Check if we should stop every second.
309     Sleeper sleeper = new Sleeper(1000, this);
310     while (!this.stopped) {
311       sleeper.sleep();
312     }
313   }
314 
315   /**
316    * Finish initialization of HMaster after becoming the primary master.
317    *
318    * <ol>
319    * <li>Initialize master components - file system manager, server manager,
320    *     assignment manager, region server tracker, catalog tracker, etc</li>
321    * <li>Start necessary service threads - rpc server, info server,
322    *     executor services, etc</li>
323    * <li>Set cluster as UP in ZooKeeper</li>
324    * <li>Wait for RegionServers to check-in</li>
325    * <li>Split logs and perform data recovery, if necessary</li>
326    * <li>Ensure assignment of root and meta regions<li>
327    * <li>Handle either fresh cluster start or master failover</li>
328    * </ol>
329    *
330    * @throws IOException
331    * @throws InterruptedException
332    * @throws KeeperException
333    */
334   private void finishInitialization()
335   throws IOException, InterruptedException, KeeperException {
336 
337     isActiveMaster = true;
338 
339     /*
340      * We are active master now... go initialize components we need to run.
341      * Note, there may be dross in zk from previous runs; it'll get addressed
342      * below after we determine if cluster startup or failover.
343      */
344 
345     // TODO: Do this using Dependency Injection, using PicoContainer, Guice or Spring.
346     this.fileSystemManager = new MasterFileSystem(this, metrics);
347     this.connection = HConnectionManager.getConnection(conf);
348     this.executorService = new ExecutorService(getServerName());
349 
350     this.serverManager = new ServerManager(this, this, metrics);
351 
352     this.catalogTracker = new CatalogTracker(this.zooKeeper, this.connection,
353       this, conf.getInt("hbase.master.catalog.timeout", Integer.MAX_VALUE));
354     this.catalogTracker.start();
355 
356     this.assignmentManager = new AssignmentManager(this, serverManager,
357       this.catalogTracker, this.executorService);
358     this.balancer = new LoadBalancer(conf);
359     zooKeeper.registerListenerFirst(assignmentManager);
360 
361     this.regionServerTracker = new RegionServerTracker(zooKeeper, this,
362       this.serverManager);
363     this.regionServerTracker.start();
364 
365     // Set the cluster as up.  If new RSs, they'll be waiting on this before
366     // going ahead with their startup.
367     this.clusterStatusTracker = new ClusterStatusTracker(getZooKeeper(), this);
368     this.clusterStatusTracker.start();
369     boolean wasUp = this.clusterStatusTracker.isClusterUp();
370     if (!wasUp) this.clusterStatusTracker.setClusterUp();
371 
372     LOG.info("Server active/primary master; " + this.address +
373       ", sessionid=0x" +
374       Long.toHexString(this.zooKeeper.getZooKeeper().getSessionId()) +
375       ", cluster-up flag was=" + wasUp);
376 
377     // start up all service threads.
378     startServiceThreads();
379 
380     // Wait for region servers to report in.  Returns count of regions.
381     int regionCount = this.serverManager.waitForRegionServers();
382 
383     // TODO: Should do this in background rather than block master startup
384     this.fileSystemManager.
385       splitLogAfterStartup(this.serverManager.getOnlineServers());
386 
387     // Make sure root and meta assigned before proceeding.
388     assignRootAndMeta();
389 
390     // Is this fresh start with no regions assigned or are we a master joining
391     // an already-running cluster?  If regionsCount == 0, then for sure a
392     // fresh start.  TOOD: Be fancier.  If regionsCount == 2, perhaps the
393     // 2 are .META. and -ROOT- and we should fall into the fresh startup
394     // branch below.  For now, do processFailover.
395     if (regionCount == 0) {
396       LOG.info("Master startup proceeding: cluster startup");
397       this.assignmentManager.cleanoutUnassigned();
398       this.assignmentManager.assignAllUserRegions();
399     } else {
400       LOG.info("Master startup proceeding: master failover");
401       this.assignmentManager.processFailover();
402     }
403 
404     // Start balancer and meta catalog janitor after meta and regions have
405     // been assigned.
406     this.balancerChore = getAndStartBalancerChore(this);
407     this.catalogJanitorChore =
408       Threads.setDaemonThreadRunning(new CatalogJanitor(this, this));
409 
410     LOG.info("Master has completed initialization");
411     initialized = true;
412   }
413 
414   /**
415    * Check <code>-ROOT-</code> and <code>.META.</code> are assigned.  If not,
416    * assign them.
417    * @throws InterruptedException
418    * @throws IOException
419    * @throws KeeperException
420    * @return Count of regions we assigned.
421    */
422   int assignRootAndMeta()
423   throws InterruptedException, IOException, KeeperException {
424     int assigned = 0;
425     long timeout = this.conf.getLong("hbase.catalog.verification.timeout", 1000);
426 
427     // Work on ROOT region.  Is it in zk in transition?
428     boolean rit = this.assignmentManager.
429       processRegionInTransitionAndBlockUntilAssigned(HRegionInfo.ROOT_REGIONINFO);
430     if (!catalogTracker.verifyRootRegionLocation(timeout)) {
431       this.assignmentManager.assignRoot();
432       this.catalogTracker.waitForRoot();
433       assigned++;
434     }
435     LOG.info("-ROOT- assigned=" + assigned + ", rit=" + rit +
436       ", location=" + catalogTracker.getRootLocation());
437 
438     // Work on meta region
439     rit = this.assignmentManager.
440       processRegionInTransitionAndBlockUntilAssigned(HRegionInfo.FIRST_META_REGIONINFO);
441     if (!this.catalogTracker.verifyMetaRegionLocation(timeout)) {
442       this.assignmentManager.assignMeta();
443       this.catalogTracker.waitForMeta();
444       // Above check waits for general meta availability but this does not
445       // guarantee that the transition has completed
446       this.assignmentManager.waitForAssignment(HRegionInfo.FIRST_META_REGIONINFO);
447       assigned++;
448     }
449     LOG.info(".META. assigned=" + assigned + ", rit=" + rit +
450       ", location=" + catalogTracker.getMetaLocation());
451     return assigned;
452   }
453 
454   /*
455    * @return This masters' address.
456    * @throws UnknownHostException
457    */
458   private static String getMyAddress(final Configuration c)
459   throws UnknownHostException {
460     // Find out our address up in DNS.
461     String s = Strings.domainNamePointerToHostName(DNS.getDefaultHost(c.get(
462         "hbase.master.dns.interface", "default"), c.get(
463         "hbase.master.dns.nameserver", "default")));
464     s += ":" + c.get(HConstants.MASTER_PORT,
465         Integer.toString(HConstants.DEFAULT_MASTER_PORT));
466     return s;
467   }
468 
469   /** @return HServerAddress of the master server */
470   public HServerAddress getMasterAddress() {
471     return this.address;
472   }
473 
474   public long getProtocolVersion(String protocol, long clientVersion) {
475     return HBaseRPCProtocolVersion.versionID;
476   }
477 
478   /** @return InfoServer object. Maybe null.*/
479   public InfoServer getInfoServer() {
480     return this.infoServer;
481   }
482 
483   @Override
484   public Configuration getConfiguration() {
485     return this.conf;
486   }
487 
488   @Override
489   public ServerManager getServerManager() {
490     return this.serverManager;
491   }
492 
493   @Override
494   public ExecutorService getExecutorService() {
495     return this.executorService;
496   }
497 
498   @Override
499   public MasterFileSystem getMasterFileSystem() {
500     return this.fileSystemManager;
501   }
502 
503   /**
504    * Get the ZK wrapper object - needed by master_jsp.java
505    * @return the zookeeper wrapper
506    */
507   public ZooKeeperWatcher getZooKeeperWatcher() {
508     return this.zooKeeper;
509   }
510 
511   /*
512    * Start up all services. If any of these threads gets an unhandled exception
513    * then they just die with a logged message.  This should be fine because
514    * in general, we do not expect the master to get such unhandled exceptions
515    *  as OOMEs; it should be lightly loaded. See what HRegionServer does if
516    *  need to install an unexpected exception handler.
517    */
518   private void startServiceThreads() throws IOException{
519  
520    // Start the executor service pools
521    this.executorService.startExecutorService(ExecutorType.MASTER_OPEN_REGION,
522       conf.getInt("hbase.master.executor.openregion.threads", 5));
523    this.executorService.startExecutorService(ExecutorType.MASTER_CLOSE_REGION,
524       conf.getInt("hbase.master.executor.closeregion.threads", 5));
525    this.executorService.startExecutorService(ExecutorType.MASTER_SERVER_OPERATIONS,
526       conf.getInt("hbase.master.executor.serverops.threads", 3));
527    this.executorService.startExecutorService(ExecutorType.MASTER_META_SERVER_OPERATIONS,
528       conf.getInt("hbase.master.executor.serverops.threads", 5));
529    
530    // We depend on there being only one instance of this executor running
531    // at a time.  To do concurrency, would need fencing of enable/disable of
532    // tables.
533    this.executorService.startExecutorService(ExecutorType.MASTER_TABLE_OPERATIONS, 1);
534 
535    // Start log cleaner thread
536    String n = Thread.currentThread().getName();
537    this.logCleaner =
538       new LogCleaner(conf.getInt("hbase.master.cleaner.interval", 60 * 1000),
539          this, conf, getMasterFileSystem().getFileSystem(),
540          getMasterFileSystem().getOldLogDir());
541          Threads.setDaemonThreadRunning(logCleaner, n + ".oldLogCleaner");
542 
543    // Put up info server.
544    int port = this.conf.getInt("hbase.master.info.port", 60010);
545    if (port >= 0) {
546      String a = this.conf.get("hbase.master.info.bindAddress", "0.0.0.0");
547      this.infoServer = new InfoServer(MASTER, a, port, false);
548      this.infoServer.setAttribute(MASTER, this);
549      this.infoServer.start();
550     }
551    
552     // Start allowing requests to happen.
553     this.rpcServer.openServer();
554     if (LOG.isDebugEnabled()) {
555       LOG.debug("Started service threads");
556     }
557 
558   }
559 
560   private void stopServiceThreads() {
561     if (LOG.isDebugEnabled()) {
562       LOG.debug("Stopping service threads");
563     }
564     if (this.rpcServer != null) this.rpcServer.stop();
565     // Clean up and close up shop
566     if (this.logCleaner!= null) this.logCleaner.interrupt();
567     if (this.infoServer != null) {
568       LOG.info("Stopping infoServer");
569       try {
570         this.infoServer.stop();
571       } catch (Exception ex) {
572         ex.printStackTrace();
573       }
574     }
575     if (this.executorService != null) this.executorService.shutdown();
576   }
577 
578   private static Thread getAndStartBalancerChore(final HMaster master) {
579     String name = master.getServerName() + "-BalancerChore";
580     int period = master.getConfiguration().getInt("hbase.balancer.period", 300000);
581     // Start up the load balancer chore
582     Chore chore = new Chore(name, period, master) {
583       @Override
584       protected void chore() {
585         master.balance();
586       }
587     };
588     return Threads.setDaemonThreadRunning(chore);
589   }
590 
591   private void stopChores() {
592     if (this.balancerChore != null) {
593       this.balancerChore.interrupt();
594     }
595     if (this.catalogJanitorChore != null) {
596       this.catalogJanitorChore.interrupt();
597     }
598   }
599 
600   @Override
601   public MapWritable regionServerStartup(final HServerInfo serverInfo,
602     final long serverCurrentTime)
603   throws IOException {
604     // Set the ip into the passed in serverInfo.  Its ip is more than likely
605     // not the ip that the master sees here.  See at end of this method where
606     // we pass it back to the regionserver by setting "hbase.regionserver.address"
607     // Everafter, the HSI combination 'server name' is what uniquely identifies
608     // the incoming RegionServer.
609     InetSocketAddress address = new InetSocketAddress(
610         HBaseServer.getRemoteIp().getHostName(),
611         serverInfo.getServerAddress().getPort());
612     serverInfo.setServerAddress(new HServerAddress(address));
613 
614     // Register with server manager
615     this.serverManager.regionServerStartup(serverInfo, serverCurrentTime);
616     // Send back some config info
617     MapWritable mw = createConfigurationSubset();
618      mw.put(new Text("hbase.regionserver.address"),
619          serverInfo.getServerAddress());
620     return mw;
621   }
622 
623   /**
624    * @return Subset of configuration to pass initializing regionservers: e.g.
625    * the filesystem to use and root directory to use.
626    */
627   protected MapWritable createConfigurationSubset() {
628     MapWritable mw = addConfig(new MapWritable(), HConstants.HBASE_DIR);
629     return addConfig(mw, "fs.default.name");
630   }
631 
632   private MapWritable addConfig(final MapWritable mw, final String key) {
633     mw.put(new Text(key), new Text(this.conf.get(key)));
634     return mw;
635   }
636 
637   @Override
638   public HMsg [] regionServerReport(HServerInfo serverInfo, HMsg msgs[],
639     HRegionInfo[] mostLoadedRegions)
640   throws IOException {
641     return adornRegionServerAnswer(serverInfo,
642       this.serverManager.regionServerReport(serverInfo, msgs, mostLoadedRegions));
643   }
644 
645   /**
646    * Override if you'd add messages to return to regionserver <code>hsi</code>
647    * or to send an exception.
648    * @param msgs Messages to add to
649    * @return Messages to return to
650    * @throws IOException exceptions that were injected for the region servers
651    */
652   protected HMsg [] adornRegionServerAnswer(final HServerInfo hsi,
653       final HMsg [] msgs) throws IOException {
654     return msgs;
655   }
656 
657   public boolean isMasterRunning() {
658     return !isStopped();
659   }
660 
661   @Override
662   public boolean balance() {
663     // If balance not true, don't run balancer.
664     if (!this.balanceSwitch) return false;
665     synchronized (this.balancer) {
666       // Only allow one balance run at at time.
667       if (this.assignmentManager.isRegionsInTransition()) {
668         LOG.debug("Not running balancer because " +
669           this.assignmentManager.getRegionsInTransition().size() +
670           " region(s) in transition: " +
671           org.apache.commons.lang.StringUtils.
672             abbreviate(this.assignmentManager.getRegionsInTransition().toString(), 256));
673         return false;
674       }
675       if (this.serverManager.areDeadServersInProgress()) {
676         LOG.debug("Not running balancer because processing dead regionserver(s): "  +
677           this.serverManager.getDeadServers());
678         return false;
679       }
680       Map<HServerInfo, List<HRegionInfo>> assignments =
681         this.assignmentManager.getAssignments();
682       // Returned Map from AM does not include mention of servers w/o assignments.
683       for (Map.Entry<String, HServerInfo> e:
684           this.serverManager.getOnlineServers().entrySet()) {
685         HServerInfo hsi = e.getValue();
686         if (!assignments.containsKey(hsi)) {
687           assignments.put(hsi, new ArrayList<HRegionInfo>());
688         }
689       }
690       List<RegionPlan> plans = this.balancer.balanceCluster(assignments);
691       if (plans != null && !plans.isEmpty()) {
692         for (RegionPlan plan: plans) {
693           LOG.info("balance " + plan);
694           this.assignmentManager.balance(plan);
695         }
696       }
697     }
698     return true;
699   }
700 
701   @Override
702   public boolean balanceSwitch(final boolean b) {
703     boolean oldValue = this.balanceSwitch;
704     this.balanceSwitch = b;
705     LOG.info("Balance=" + b);
706     return oldValue;
707   }
708 
709   /**
710    * Switch for the background {@link CatalogJanitor} thread.
711    * Used for testing.  The thread will continue to run.  It will just be a noop
712    * if disabled.
713    * @param b If false, the catalog janitor won't do anything.
714    */
715   public void setCatalogJanitorEnabled(final boolean b) {
716     ((CatalogJanitor)this.catalogJanitorChore).setEnabled(b);
717   }
718 
719   @Override
720   public void move(final byte[] encodedRegionName, final byte[] destServerName)
721   throws UnknownRegionException {
722     Pair<HRegionInfo, HServerInfo> p =
723       this.assignmentManager.getAssignment(encodedRegionName);
724     if (p == null)
725       throw new UnknownRegionException(Bytes.toStringBinary(encodedRegionName));
726     HRegionInfo hri = p.getFirst();
727     HServerInfo dest = null;
728     if (destServerName == null || destServerName.length == 0) {
729       LOG.info("Passed destination servername is null/empty so " +
730         "choosing a server at random");
731       this.assignmentManager.clearRegionPlan(hri);
732       // Unassign will reassign it elsewhere choosing random server.
733       this.assignmentManager.unassign(hri);
734     } else {
735       dest = this.serverManager.getServerInfo(new String(destServerName));
736       RegionPlan rp = new RegionPlan(p.getFirst(), p.getSecond(), dest);
737       this.assignmentManager.balance(rp);
738     }
739   }
740 
741   public void createTable(HTableDescriptor desc, byte [][] splitKeys)
742   throws IOException {
743     createTable(desc, splitKeys, false);
744   }
745 
746   public void createTable(HTableDescriptor desc, byte [][] splitKeys,
747       boolean sync)
748   throws IOException {
749     if (!isMasterRunning()) {
750       throw new MasterNotRunningException();
751     }
752     HRegionInfo [] newRegions = null;
753     if(splitKeys == null || splitKeys.length == 0) {
754       newRegions = new HRegionInfo [] { new HRegionInfo(desc, null, null) };
755     } else {
756       int numRegions = splitKeys.length + 1;
757       newRegions = new HRegionInfo[numRegions];
758       byte [] startKey = null;
759       byte [] endKey = null;
760       for(int i=0;i<numRegions;i++) {
761         endKey = (i == splitKeys.length) ? null : splitKeys[i];
762         newRegions[i] = new HRegionInfo(desc, startKey, endKey);
763         startKey = endKey;
764       }
765     }
766     int timeout = conf.getInt("hbase.client.catalog.timeout", 10000);
767     // Need META availability to create a table
768     try {
769       if(catalogTracker.waitForMeta(timeout) == null) {
770         throw new NotAllMetaRegionsOnlineException();
771       }
772     } catch (InterruptedException e) {
773       LOG.warn("Interrupted waiting for meta availability", e);
774       throw new IOException(e);
775     }
776     createTable(newRegions, sync);
777   }
778 
779   private synchronized void createTable(final HRegionInfo [] newRegions,
780       final boolean sync)
781   throws IOException {
782     String tableName = newRegions[0].getTableDesc().getNameAsString();
783     if(MetaReader.tableExists(catalogTracker, tableName)) {
784       throw new TableExistsException(tableName);
785     }
786     for (HRegionInfo newRegion : newRegions) {
787       // 1. Set table enabling flag up in zk.
788       try {
789         assignmentManager.getZKTable().setEnabledTable(tableName);
790       } catch (KeeperException e) {
791         throw new IOException("Unable to ensure that the table will be" +
792           " enabled because of a ZooKeeper issue", e);
793       }
794 
795       // 2. Create HRegion
796       HRegion region = HRegion.createHRegion(newRegion,
797         fileSystemManager.getRootDir(), conf);
798 
799       // 3. Insert into META
800       MetaEditor.addRegionToMeta(catalogTracker, region.getRegionInfo());
801 
802       // 4. Close the new region to flush to disk.  Close log file too.
803       region.close();
804       region.getLog().closeAndDelete();
805     }
806 
807     // 5. Trigger immediate assignment of the regions in round-robin fashion
808     if (newRegions.length == 1) {
809       this.assignmentManager.assign(newRegions[0], true);
810     } else {
811       List<HServerInfo> servers = serverManager.getOnlineServersList();
812       this.assignmentManager.bulkAssignUserRegions(newRegions, servers, sync);
813     }
814 
815     // 6. If sync, wait for assignment of regions
816     if (sync) {
817       LOG.debug("Waiting for " + newRegions.length + " region(s) to be assigned");
818       for (HRegionInfo regionInfo : newRegions) {
819         try {
820           this.assignmentManager.waitForAssignment(regionInfo);
821         } catch (InterruptedException e) {
822           LOG.info("Interrupted waiting for region to be assigned during " +
823               "create table call", e);
824           Thread.currentThread().interrupt();
825           return;
826         }
827       }
828     }
829   }
830 
831   private static boolean isCatalogTable(final byte [] tableName) {
832     return Bytes.equals(tableName, HConstants.ROOT_TABLE_NAME) ||
833            Bytes.equals(tableName, HConstants.META_TABLE_NAME);
834   }
835 
836   public void deleteTable(final byte [] tableName) throws IOException {
837     this.executorService.submit(new DeleteTableHandler(tableName, this, this));
838   }
839 
840   public void addColumn(byte [] tableName, HColumnDescriptor column)
841   throws IOException {
842     new TableAddFamilyHandler(tableName, column, this, this).process();
843   }
844 
845   public void modifyColumn(byte [] tableName, HColumnDescriptor descriptor)
846   throws IOException {
847     new TableModifyFamilyHandler(tableName, descriptor, this, this).process();
848   }
849 
850   public void deleteColumn(final byte [] tableName, final byte [] c)
851   throws IOException {
852     new TableDeleteFamilyHandler(tableName, c, this, this).process();
853   }
854 
855   public void enableTable(final byte [] tableName) throws IOException {
856     this.executorService.submit(new EnableTableHandler(this, tableName,
857       catalogTracker, assignmentManager));
858   }
859 
860   public void disableTable(final byte [] tableName) throws IOException {
861     this.executorService.submit(new DisableTableHandler(this, tableName,
862       catalogTracker, assignmentManager));
863   }
864 
865   /**
866    * Return the region and current deployment for the region containing
867    * the given row. If the region cannot be found, returns null. If it
868    * is found, but not currently deployed, the second element of the pair
869    * may be null.
870    */
871   Pair<HRegionInfo,HServerAddress> getTableRegionForRow(
872       final byte [] tableName, final byte [] rowKey)
873   throws IOException {
874     final AtomicReference<Pair<HRegionInfo, HServerAddress>> result =
875       new AtomicReference<Pair<HRegionInfo, HServerAddress>>(null);
876 
877     MetaScannerVisitor visitor =
878       new MetaScannerVisitor() {
879         @Override
880         public boolean processRow(Result data) throws IOException {
881           if (data == null || data.size() <= 0) {
882             return true;
883           }
884           Pair<HRegionInfo, HServerAddress> pair =
885             MetaReader.metaRowToRegionPair(data);
886           if (pair == null) {
887             return false;
888           }
889           if (!Bytes.equals(pair.getFirst().getTableDesc().getName(),
890                 tableName)) {
891             return false;
892           }
893           result.set(pair);
894           return true;
895         }
896     };
897 
898     MetaScanner.metaScan(conf, visitor, tableName, rowKey, 1);
899     return result.get();
900   }
901 
902   @Override
903   public void modifyTable(final byte[] tableName, HTableDescriptor htd)
904   throws IOException {
905     this.executorService.submit(new ModifyTableHandler(tableName, htd, this, this));
906   }
907 
908   @Override
909   public void checkTableModifiable(final byte [] tableName)
910   throws IOException {
911     String tableNameStr = Bytes.toString(tableName);
912     if (isCatalogTable(tableName)) {
913       throw new IOException("Can't modify catalog tables");
914     }
915     if (!MetaReader.tableExists(getCatalogTracker(), tableNameStr)) {
916       throw new TableNotFoundException(tableNameStr);
917     }
918     if (!getAssignmentManager().getZKTable().
919         isDisabledTable(Bytes.toString(tableName))) {
920       throw new TableNotDisabledException(tableName);
921     }
922   }
923 
924   public void clearFromTransition(HRegionInfo hri) {
925     if (this.assignmentManager.isRegionInTransition(hri) != null) {
926       this.assignmentManager.clearRegionFromTransition(hri);
927     }
928   }
929   /**
930    * @return cluster status
931    */
932   public ClusterStatus getClusterStatus() {
933     ClusterStatus status = new ClusterStatus();
934     status.setHBaseVersion(VersionInfo.getVersion());
935     status.setServerInfo(serverManager.getOnlineServers().values());
936     status.setDeadServers(serverManager.getDeadServers());
937     status.setRegionsInTransition(assignmentManager.getRegionsInTransition());
938     return status;
939   }
940 
941   @Override
942   public void abort(final String msg, final Throwable t) {
943     if (t != null) LOG.fatal(msg, t);
944     else LOG.fatal(msg);
945     this.abort = true;
946     stop("Aborting");
947   }
948 
949   @Override
950   public ZooKeeperWatcher getZooKeeper() {
951     return zooKeeper;
952   }
953 
954   @Override
955   public String getServerName() {
956     return address.toString();
957   }
958 
959   @Override
960   public CatalogTracker getCatalogTracker() {
961     return catalogTracker;
962   }
963 
964   @Override
965   public AssignmentManager getAssignmentManager() {
966     return this.assignmentManager;
967   }
968 
969   @Override
970   public void shutdown() {
971     this.serverManager.shutdownCluster();
972     try {
973       this.clusterStatusTracker.setClusterDown();
974     } catch (KeeperException e) {
975       LOG.error("ZooKeeper exception trying to set cluster as down in ZK", e);
976     }
977   }
978 
979   @Override
980   public void stopMaster() {
981     stop("Stopped by " + Thread.currentThread().getName());
982   }
983 
984   @Override
985   public void stop(final String why) {
986     LOG.info(why);
987     this.stopped = true;
988     // If we are a backup master, we need to interrupt wait
989     synchronized (this.activeMasterManager.clusterHasActiveMaster) {
990       this.activeMasterManager.clusterHasActiveMaster.notifyAll();
991     }
992   }
993 
994   @Override
995   public boolean isStopped() {
996     return this.stopped;
997   }
998 
999   /**
1000    * Report whether this master is currently the active master or not.
1001    * If not active master, we are parked on ZK waiting to become active.
1002    *
1003    * This method is used for testing.
1004    *
1005    * @return true if active master, false if not.
1006    */
1007   public boolean isActiveMaster() {
1008     return isActiveMaster;
1009   }
1010 
1011   /**
1012    * Report whether this master has completed with its initialization and is
1013    * ready.  If ready, the master is also the active master.  A standby master
1014    * is never ready.
1015    *
1016    * This method is used for testing.
1017    *
1018    * @return true if master is ready to go, false if not.
1019    */
1020   public boolean isInitialized() {
1021     return initialized;
1022   }
1023 
1024   @Override
1025   public void assign(final byte [] regionName, final boolean force)
1026   throws IOException {
1027     Pair<HRegionInfo, HServerAddress> pair =
1028       MetaReader.getRegion(this.catalogTracker, regionName);
1029     if (pair == null) throw new UnknownRegionException(Bytes.toString(regionName));
1030     assignRegion(pair.getFirst());
1031   }
1032 
1033   public void assignRegion(HRegionInfo hri) {
1034     assignmentManager.assign(hri, true);
1035   }
1036 
1037   @Override
1038   public void unassign(final byte [] regionName, final boolean force)
1039   throws IOException {
1040     Pair<HRegionInfo, HServerAddress> pair =
1041       MetaReader.getRegion(this.catalogTracker, regionName);
1042     if (pair == null) throw new UnknownRegionException(Bytes.toStringBinary(regionName));
1043     HRegionInfo hri = pair.getFirst();
1044     if (force) this.assignmentManager.clearRegionFromTransition(hri);
1045     this.assignmentManager.unassign(hri, force);
1046   }
1047 
1048   /**
1049    * Utility for constructing an instance of the passed HMaster class.
1050    * @param masterClass
1051    * @param conf
1052    * @return HMaster instance.
1053    */
1054   public static HMaster constructMaster(Class<? extends HMaster> masterClass,
1055       final Configuration conf)  {
1056     try {
1057       Constructor<? extends HMaster> c =
1058         masterClass.getConstructor(Configuration.class);
1059       return c.newInstance(conf);
1060     } catch (InvocationTargetException ite) {
1061       Throwable target = ite.getTargetException() != null?
1062         ite.getTargetException(): ite;
1063       if (target.getCause() != null) target = target.getCause();
1064       throw new RuntimeException("Failed construction of Master: " +
1065         masterClass.toString(), target);
1066     } catch (Exception e) {
1067       throw new RuntimeException("Failed construction of Master: " +
1068         masterClass.toString() + ((e.getCause() != null)?
1069           e.getCause().getMessage(): ""), e);
1070     }
1071   }
1072 
1073 
1074   /**
1075    * @see org.apache.hadoop.hbase.master.HMasterCommandLine
1076    */
1077   public static void main(String [] args) throws Exception {
1078     new HMasterCommandLine(HMaster.class).doMain(args);
1079   }
1080 }