Hadoop2.6.0的事件分类与实现

标签：

前言

说实在的，在阅读Hadoop YARN的源码之前，我对于java枚举的使用相形见绌。YARN中实现的事件在可读性、可维护性、可扩展性方面的工作都值得借鉴。

概念

在具体分析源码之前，我们先看看YARN是如何定义一个事件的。比如作业启动的事件，很多人可能会用常量将它定义到一个class文件中，就像下面这样：

class Constants {
  public static final String JOB_START_EVENT = "jobStart";
}

enum Enums {
  JOB_START_EVENT("jobStart");
  private String name;
  private Enums(String name) {
    this.name = name;
  }
}

class Constants {
  public static final String JOB_START_EVENT = "jobStart";
  public static final String JOB_END_EVENT = "jobEnd";
}

enum Enums {
  JOB_START_EVENT("jobStart"),
  JOB_END_EVENT("jobEnd");
  private String name;
  private Enums(String name) {
    this.name = name;
  }
}

class Constants {
  public static final String JOB_START_EVENT = "jobStart";
  public static final String JOB_END_EVENT = "jobEnd";
  public static final String TASK_START_EVENT = "taskStart";
  public static final String TASK_END_EVENT = "taskEnd";
  // 其它各种概念的常量
}

enum Enums {
  JOB_START_EVENT("jobStart"),
  JOB_END_EVENT("jobEnd"),
  // 其它各种概念的常量枚举
  TASK_START_EVENT("taskStart"),
  TASK_END_EVENT("taskEnd");
  private String name;
  private Enums(String name) {
    this.name = name;
  }
}

class JobConstants {
  public static final String JOB_START_EVENT = "jobStart";
  public static final String JOB_END_EVENT = "jobEnd";
}

class TaskConstants {
  public static final String TASK_START_EVENT = "taskStart";
  public static final String TASK_END_EVENT = "taskEnd";
}

enum JobEnums {
  JOB_START_EVENT("jobStart"),
  JOB_END_EVENT("jobEnd");
  private String name;
  private JobEnums (String name) {
    this.name = name;
  }
}

enum TaskEnums {
  TASK_START_EVENT("taskStart"),
  TASK_END_EVENT("taskEnd");
  private String name;
  private TaskEnums (String name) {
    this.name = name;
  }
}

enum JobEnums {
  JOB_START_EVENT(10, "jobStart"),
  JOB_END_EVENT(20, "jobEnd");
  private int code;
  private String name;
  private JobEnums (int code, String name) {
    this.code = code;  
    this.name = name;
  }
}

enum TaskEnums {
  TASK_START_EVENT(110, "taskStart"),
  TASK_END_EVENT(120, "taskEnd");
  private int code;
  private String name;
  private TaskEnums (int code, String name)   {
    this.code = code;
    this.name = name;
  }
}

enum JobEnums {
  JOB_START_EVENT(10, "jobStart", "job start description"),
  JOB_END_EVENT(20, "jobEnd", "job end description");
  private int code;
  private String name;
  private String description;
  private JobEnums (int code, String name, String description) {
    this.code = code;  
    this.name = name;
    this.description = description;
  }

  public int hashCode() {
    return this.name.hashCode() + this.description.hashCode();
  }
}

enum TaskEnums {
  TASK_START_EVENT(110, "taskStart", 1460977775087),
  TASK_END_EVENT(120, "taskEnd", 1460977775088);
  private int code;
  private String name;
  private long timestamp;
  private TaskEnums (int code, String name, long timestamp)   {
    this.code = code;
    this.name = name;
    this.timestamp = timestamp;
  }

  public int hashCode() {
    return this.name.hashCode();
  }
}

事件 = 事件名称 + 事件类型

比如作业启动事件 = 作业事件 + 作业事件类型

事件与事件类型

Hadoop2.6.0中的事件多种多样，最为常见的包括：ContainerEvent、ApplicationEvent、JobEvent、RMAppEvent、RMAppAttemptEvent、TaskEvent、TaskAttemptEvent等。为了解决枚举与常量在可读性、可维护性、可复用性、可扩展性等方面的问题，Hadoop对事件进行了以下抽象：

/**
 * Interface defining events api.
 *
 */
@Public
@Evolving
public interface Event<TYPE extends Enum<TYPE>> {

  TYPE getType();
  long getTimestamp();
  String toString();
}

以上接口说明了任何一个具体事件都是一个枚举类型，而且有一个事件类型属性（用泛型标记TYPE表示），一个时间戳及toString()方法。

所有事件都有一个基本实现AbstractEvent，其实现如下：

/**
 * Parent class of all the events. All events extend this class.
 */
@Public
@Evolving
public abstract class AbstractEvent<TYPE extends Enum<TYPE>> 
    implements Event<TYPE> {

  private final TYPE type;
  private final long timestamp;

  // use this if you DON'T care about the timestamp
  public AbstractEvent(TYPE type) {
    this.type = type;
    // We're not generating a real timestamp here.  It's too expensive.
    timestamp = -1L;
  }

  // use this if you care about the timestamp
  public AbstractEvent(TYPE type, long timestamp) {
    this.type = type;
    this.timestamp = timestamp;
  }

  @Override
  public long getTimestamp() {
    return timestamp;
  }

  @Override
  public TYPE getType() {
    return type;
  }

  @Override
  public String toString() {
    return "EventType: " + getType();
  }
}

/**
 * This class encapsulates job related events.
 *
 */
public class JobEvent extends AbstractEvent<JobEventType> {

  private JobId jobID;

  public JobEvent(JobId jobID, JobEventType type) {
    super(type);
    this.jobID = jobID;
  }

  public JobId getJobId() {
    return jobID;
  }

}

/**
 * this class encapsulates task related events.
 *
 */
public class TaskEvent extends AbstractEvent<TaskEventType> {

  private TaskId taskID;

  public TaskEvent(TaskId taskID, TaskEventType type) {
    super(type);
    this.taskID = taskID;
  }

  public TaskId getTaskID() {
    return taskID;
  }
}

事件类型属性（用泛型标记TYPE表示）在任务事件中对应的是TaskEventType，其实现如下：

/**
 * Event types handled by Task.
 */
public enum TaskEventType {

  //Producer:Client, Job
  T_KILL,

  //Producer:Job
  T_SCHEDULE,
  T_RECOVER,

  //Producer:Speculator
  T_ADD_SPEC_ATTEMPT,

  //Producer:TaskAttempt
  T_ATTEMPT_LAUNCHED,
  T_ATTEMPT_COMMIT_PENDING,
  T_ATTEMPT_FAILED,
  T_ATTEMPT_SUCCEEDED,
  T_ATTEMPT_KILLED
}

这种实现将枚举与各种事件之间的差异（表现在属性和方法的不同）解耦，极大地扩展了可读性、可维护性，并且保留了相同逻辑的代码复用。

后记：个人总结整理的《深入理解Spark：核心思想与源码分析》一书现在已经正式出版上市，目前京东、当当、天猫等网站均有销售，欢迎感兴趣的同学购买。

京东(现有满150减50活动）)：http://item.jd.com/11846120.html 

当当：http://product.dangdang.com/23838168.html 

Hadoop2.6.0的事件分类与实现

标签：

原文地址：http://blog.csdn.net/beliefer/article/details/51180261