HBase源码系列（三）HMaster启动过程

这是HBase server端开始的第一章。

在HMaster的构造函数，从方法注释上看到，主要做了两件事情：
1、初始化本地的HRegionServer；
2、启动ActiveMasterManager；

public HMaster(final Configuration conf, CoordinatedStateManager csm)
        throws IOException, KeeperException, InterruptedException {
    super(conf, csm);
    ...
    this.metricsMaster = new MetricsMaster(new MetricsMasterWrapperImpl(this));

    ...
    // Some unit tests don't need a cluster, so no zookeeper at all
    if (!conf.getBoolean("hbase.testing.nocluster", false)) {
        activeMasterManager = new ActiveMasterManager(zooKeeper, this.serverName, this);
        int infoPort = putUpJettyServer();
        startActiveMasterManager(infoPort);
    } else {
        activeMasterManager = null;
    }
}

HMaster是HRegionServer的子类，HRegionServer又是抽象类HasThread的子类，HasThread实现了Runnable接口。

HRegionServer的构造中，启动rpc通信。

public HRegionServer(Configuration conf, CoordinatedStateManager csm) throws IOException, InterruptedException {
    rpcServices = createRpcServices();
    rpcServices.start();
}

/** Starts the service.  Must be called before any calls will be handled. */
@Override
public synchronized void start() {
  if (started) return;
  authTokenSecretMgr = createSecretManager();
  if (authTokenSecretMgr != null) {
    setSecretManager(authTokenSecretMgr);
    authTokenSecretMgr.start();
  }
  this.authManager = new ServiceAuthorizationManager();
  HBasePolicyProvider.init(conf, authManager);
  responder.start();
  listener.start();
  scheduler.start();
  started = true;
}

在启动了HMaster之后，从finishActiveMasterInitialization（MonitoredTask status）方法的注释上，看到完成HMaster的初始化过程做了许多工作：
1、初始化master的组件，如file system manager, server manager,assignment manager, region server tracker等；
2、启动必要的服务线程，如balancer, catalog janior,executor services等；
3、等待RegionServers注册；
4、切割日志，并进行数据恢复（如果需要）；
5、分配 meta/namespace regions；

阻塞等待ActiveMaster启动完成：

public void run() {
try {
    while (!master.isStopped() && master.isActiveMaster()) {
        Thread.sleep(timeout);
        if (master.isInitialized()) {
            LOG.debug("Initialization completed within allotted tolerance. Monitor exiting.");
        } else {
            LOG.error("Master failed to complete initialization after " + timeout + "ms. Please"
                    + " consider submitting a bug report including a thread dump of this process.");
            if (haltOnTimeout) {
                LOG.error("Zombie Master exiting. Thread dump to stdout");
                Threads.printThreadInfo(System.out, "Zombie HMaster");
                System.exit(-1);
            }
        }
    }
} catch (InterruptedException ie) {
    LOG.trace("InitMonitor thread interrupted. Existing.");
}
}

HMaster初始化完成后，后面才是重头戏，简单列下都实例化了哪些东西。

// file system manager for the master FS operations
private MasterFileSystem fileSystemManager;
// server manager to deal with region server info
volatile ServerManager serverManager;
    /**
 * Cluster connection to be shared by services.
 * Initialized at server startup and closed when server shuts down.
 * Clients must never close it explicitly.
 */
protected ClusterConnection clusterConnection;
    /*
 * Long-living meta table locator, which is created when the server is started and stopped
 * when server shuts down. References to this locator shall be used to perform according
 * operations in EventHandlers. Primary reason for this decision is to make it mockable
 * for tests.
 */
protected MetaTableLocator metaTableLocator;

MasterCoprocessorHost cpHost;

private TableNamespaceManager tableNamespaceManager;

...

这些类都会被实例化。开学啦，等待HRegionServer报道：

// Wait for region servers to report in
this.serverManager.waitForRegionServers(status);

分配hbase:meta表前，先split meta表的日志：
先从ZNode上拿到meta region的状态信息。
切割WAL日志：

long splitTime = 0, splitLogSize = 0;
List<Path> logDirs = getLogDirs(serverNames);

splitLogManager.handleDeadWorkers(serverNames);
splitTime = EnvironmentEdgeManager.currentTime();
splitLogSize = splitLogManager.splitLogDistributed(serverNames, logDirs, filter);
splitTime = EnvironmentEdgeManager.currentTime() - splitTime;

if (this.metricsMasterFilesystem != null) {
  if (filter == META_FILTER) {
    this.metricsMasterFilesystem.addMetaWALSplit(splitTime, splitLogSize);
  } else {
    this.metricsMasterFilesystem.addSplit(splitTime, splitLogSize);
  }
}

分配meta表
…
发送RPC到指定的server

OpenRegionResponse response = admin.openRegion(null, request);

构建user regions

this.assignmentManager.joinCluster();

分配region的工作是由assignmentManager完成的，joinCluster()方法调用processDeadServersAndRegionsInTransition（）方法的最后调用assignAllUserRegions（）

if (!failover) {
  // Fresh cluster startup.
  LOG.info("Clean cluster startup. Assigning user regions");
  assignAllUserRegions(allRegions);
}

分配用户region的方法和分配meta表的方法基本一致。

至此，启动HMaster的过程基本完成了。