【Hadoop】【JHS】1-JHS的初始化过程及其服务介绍

JobHistoryServer是一个独立的进程，所以首先看下它的main方法。
JobHistoryServer.main()

public static void main(String[] args) {
  LOG.info("The JobHistoryServer will to start");
  launchJobHistoryServer(args);
  LOG.info("The JobHistoryServer has started");
}

JobHistoryServer.launcherJobHistoryServer()

static JobHistoryServer launchJobHistoryServer(String[] args) {
  //设置未捕获异常的处理器
  Thread.setDefaultUncaughtExceptionHandler(new YarnUncaughtExceptionHandler());
  //打印启停日志
  StringUtils.startupShutdownMessage(JobHistoryServer.class, args, LOG);
  JobHistoryServer jobHistoryServer = null;
  try {
    jobHistoryServer = new JobHistoryServer();
    ShutdownHookManager.get().addShutdownHook(
        new CompositeServiceShutdownHook(jobHistoryServer),
        SHUTDOWN_HOOK_PRIORITY);
    YarnConfiguration conf = new YarnConfiguration(new JobConf());
    //解析命令行侧参数，将参数写入conf中
    new GenericOptionsParser(conf, args);
    jobHistoryServer.init(conf);
    jobHistoryServer.start();
    HSAuditLogger.logSuccess(getUserName(), "jhsStartup", "JobHistoryServer");
  } catch (Throwable t) {
    LOG.error("Error starting JobHistoryServer", t);
    HSAuditLogger.logFailure(getUserName(), "jhsStartup", "",
        "JobHistoryServer", "Exception during startup");
    ExitUtil.terminate(-1, "Error starting JobHistoryServer");
  }
  return jobHistoryServer;
}

上述的启动流程主要做了下面几件事情：
1-添加shudown hook；
2-解析客户端参数；
3-init，start
serviceStart()方法中沒有做任何额外的操作，仅仅是将添加的service全部启动。下面主要介绍下serviceInit()的过程。代码如下：

protected void serviceInit(Configuration conf) throws Exception {
  Configuration config = new YarnConfiguration(conf);

  //初始化MRWebAppUtil的http policy(HTTPS或者HTTP)
  MRWebAppUtil.initialize(getConfig());
  try {
    //安全模式下需要使用mapred principal的keytab登录，mapreduce.jobhistory.keytab配置了keytab路径；
    doSecureLogin(conf);
  } catch(IOException ie) {
    throw new YarnRuntimeException("History Server Failed to login", ie);
  }
  //1-JobHistory类用来加载和管理 Job history 缓存.
  jobHistoryService = new JobHistory();
  historyContext = (HistoryContext)jobHistoryService;
  //2-是否打开historyserver 在启动时存储和恢复服务器状态功能，可存储用来生成token的master key以及生成的DT
  stateStore = createStateStore(conf);
  //3-创建jhs中token的master key管理器
  this.jhsDTSecretManager = createJHSSecretManager(conf, stateStore);
  //4-创建jhs 客户端服务，该服务用来处理客户端JobClient请求。例如,常用的getCounters()，getJobReport()。后面会详细介绍；
  clientService = createHistoryClientService();
  //5-创建归集日志删除服务，周期性的删除已经过期聚合日志信息（Application粒度）；将/tmp/logs/下面存活周期超过：log-aggregation.retain-seconds的文件删掉的归集日志删掉
  aggLogDelService = new AggregatedLogDeletionService();
  //6-创建日志归档服务，周期性归档日志信息（Application粒度）；将nodemanager上传到/tmp/logs下面的日志归档到/tmp/archived下面，注意日志归档和日志归集的不同
  aggLogArchiveService = new AggregatedLogArchiveService();
  //7-
  hsAdminServer = new HSAdminServer(aggLogDelService, aggLogArchiveService,jobHistoryService);
  addService(stateStore);
  addService(new HistoryServerSecretManagerService());
  addService(jobHistoryService);
  addService(clientService);
  addService(aggLogDelService);
  addService(aggLogArchiveService);
  addService(hsAdminServer);

  DefaultMetricsSystem.initialize("JobHistoryServer");
  JvmMetrics jm = JvmMetrics.initSingleton("JobHistoryServer", null);
  pauseMonitor = new JvmPauseMonitor();
  addService(pauseMonitor);
  jm.setPauseMonitor(pauseMonitor);

  super.serviceInit(config);
}

上面的init代码中主要做了创建了七种不同的service，JobHistoryServer的所有功能都在这里面实现。
关于这五个service各自的功能将在其他的章节中详细介绍。