5、Nacos服务注册服务端源码分析（四）之NotifyCenter

上篇我们讲server端处理服务注册源码时，遇到了一个关键类NotifyCenter，本篇就主要来分析下这个类。

NotifyCenter

这个类所在包：nacos-common

/**
 * Unified Event Notify Center.
 */

通过类注释可以看出来这个类是一个统一的事件通知中心，那也就足可见这个类的重要性了。

public class NotifyCenter

通过类定义发现这个类就是一个普通类，不会被注册为spring bean且类中的方法都是静态方法，也就是说这个类是整个进程共享的了，所以其中的内部变量必然要是线程安全的，结合源码看看是如何处理的。

private static final Logger LOGGER = LoggerFactory.getLogger(NotifyCenter.class);
//这两个变量是在静态块中初始化的，所以不需要特殊处理
public static int ringBufferSize;
public static int shareBufferSize;

private static final AtomicBoolean CLOSED = new AtomicBoolean(false);
//初始化EventPublisher的工厂
private static final EventPublisherFactory DEFAULT_PUBLISHER_FACTORY;

private static final NotifyCenter INSTANCE = new NotifyCenter();

private DefaultSharePublisher sharePublisher;
//SPI机制获取到EventPublisher实现之后，获取class,赋值到此
private static Class<? extends EventPublisher> clazz;

/**
 * Publisher management container.
 */
private final Map<String, EventPublisher> publisherMap = new ConcurrentHashMap<>(16);

static {
    // Internal ArrayBlockingQueue buffer size. For applications with high write throughput,
    // this value needs to be increased appropriately. default value is 16384
    String ringBufferSizeProperty = "nacos.core.notify.ring-buffer-size";
    ringBufferSize = Integer.getInteger(ringBufferSizeProperty, 16384);
    
    // The size of the public publisher's message staging queue buffer
    String shareBufferSizeProperty = "nacos.core.notify.share-buffer-size";
    shareBufferSize = Integer.getInteger(shareBufferSizeProperty, 1024);
    //采用SPI拓展机制，NacosServiceLoader内部封装了JDK的ServiceLoader,在此基础上加了一层缓存
    final Collection<EventPublisher> publishers = NacosServiceLoader.load(EventPublisher.class);
    Iterator<EventPublisher> iterator = publishers.iterator();
    if (iterator.hasNext()) {
        clazz = iterator.next().getClass();
    } else {
    	//默认使用DefaultPublisher
        clazz = DefaultPublisher.class;
    }
    
    DEFAULT_PUBLISHER_FACTORY = (cls, buffer) -> {
        try {
        	//实例化EventPublisher
            EventPublisher publisher = clazz.newInstance();
            //调用EventPublisher初始化方法
            publisher.init(cls, buffer);
            return publisher;
        } catch (Throwable ex) {
            LOGGER.error("Service class newInstance has error : ", ex);
            throw new NacosRuntimeException(SERVER_ERROR, ex);
        }
    };
    
    try {
        
        // Create and init DefaultSharePublisher instance.
        INSTANCE.sharePublisher = new DefaultSharePublisher();
        INSTANCE.sharePublisher.init(SlowEvent.class, shareBufferSize);
        
    } catch (Throwable ex) {
        LOGGER.error("Service class newInstance has error : ", ex);
    }
    //JVM销毁前执行的钩子函数 NotifyCenter::shutdown
    ThreadUtils.addShutdownHook(NotifyCenter::shutdown);
}

查看代码可以知道关键的处理逻辑是：创建EventPublisher。

---

EventPublisher

以上涉及到两个EventPublisher一个是默认的DefaultPublisher，另一个是DefaultSharePublisher

我们来一个个梳理下UML中的类：

interface EventPublisher

事件通知的接口，定义了事件通知的一系列方法

/**
  * Initializes the event publisher.
  * 初始化
  * @param type       {@link Event >}
  * @param bufferSize 消息队列大小
  */
 void init(Class<? extends Event> type, int bufferSize);
 
 /**
  * The number of currently staged events.
  * 当前暂存的事件数量
  * @return event size
  */
 long currentEventSize();
 
 /**
  * Add listener.
  *	添加订阅者
  * @param subscriber {@link Subscriber}
  */
 void addSubscriber(Subscriber subscriber);
 
 /**
  * Remove listener.
  * 移除订阅者
  * @param subscriber {@link Subscriber}
  */
 void removeSubscriber(Subscriber subscriber);
 
 /**
  * publish event.
  *	发布事件
  * @param event {@link Event}
  * @return publish event is success
  */
 boolean publish(Event event);
 
 /**
  * Notify listener.
  * 通知订阅者
  * @param subscriber {@link Subscriber}
  * @param event      {@link Event}
  */
 void notifySubscriber(Subscriber subscriber, Event event);

class DefaultPublisher

public class DefaultPublisher extends Thread implements EventPublisher

DefaultPublisher 是事件发布接口EventPublisher的默认实现，除此之外，还继承了Thread，那么也就意味着DefaultPublisher除了有事件处理能力，本身还是一个线程。

先看接口中定义的初始化方法init是如何实现的，从这里看起会很好的帮助我们理解这个类：

@Override
public void init(Class<? extends Event> type, int bufferSize) {
	//设置为守护线程
    setDaemon(true);
    //设置线程名称，一定程度上这可以帮助我们排查问题
    setName("nacos.publisher-" + type.getName());
    //Event类型
    this.eventType = type;
    //队列长度
    this.queueMaxSize = bufferSize;
    //队列
    this.queue = new ArrayBlockingQueue<>(bufferSize);
    //Thread的start()方法
    start();
}
//全局变量，用来标识当前Publisher是否初始化完成
private volatile boolean initialized = false;
@Override
public synchronized void start() {
    if (!initialized) {
        // start just called once
        super.start();
        if (queueMaxSize == -1) {
            queueMaxSize = ringBufferSize;
        }
        initialized = true;
    }
}

即然继承了Thread那么就肯定回去重写关键的run()方法，我们看下是如何重写的：

@Override
public void run() {
    openEventHandler();
}

void openEventHandler() {
    try {
        
        // This variable is defined to resolve the problem which message overstock in the queue.
        // 这个变量用于解决队列中消息积压的问题
        int waitTimes = 60;
        // To ensure that messages are not lost, enable EventHandler when
        // waiting for the first Subscriber to register
        // 没有关闭 && 没有订阅者 && waitTimes>0 --> 休眠1s,最多可以sleep 60s
        while (!shutdown && !hasSubscriber() && waitTimes > 0) {
            ThreadUtils.sleep(1000L);
            waitTimes--;
        }

        while (!shutdown) {
        	//如果没有获取到事件，这里会阻塞
            final Event event = queue.take();
            //处理事件
            receiveEvent(event);
            UPDATER.compareAndSet(this, lastEventSequence, Math.max(lastEventSequence, event.sequence()));
        }
    } catch (Throwable ex) {
        LOGGER.error("Event listener exception : ", ex);
    }
}

可以看到，这里有一个死循环，只要没有收到shutdown事件，那么就会一直尝试从队列中获取事件去处理。这里获取事件的逻辑是从阻塞队列中取数据，所以当没有时间要处理时，当前线程会阻塞在获取事件的逻辑中，不会对cpu有过多的压力。

/**
 * Receive and notifySubscriber to process the event.
 *
 * @param event {@link Event}.
 */
void receiveEvent(Event event) {
    final long currentEventSequence = event.sequence();
    
    if (!hasSubscriber()) {
        LOGGER.warn("[NotifyCenter] the {} is lost, because there is no subscriber.", event);
        return;
    }
    
    // Notification single event listener
    for (Subscriber subscriber : subscribers) {
        if (!subscriber.scopeMatches(event)) {
            continue;
        }
        
        // Whether to ignore expiration events
        if (subscriber.ignoreExpireEvent() && lastEventSequence > currentEventSequence) {
            LOGGER.debug("[NotifyCenter] the {} is unacceptable to this subscriber, because had expire",
                    event.getClass());
            continue;
        }
        
        // Because unifying smartSubscriber and subscriber, so here need to think of compatibility.
        // Remove original judge part of codes.
        notifySubscriber(subscriber, event);
    }
}

@Override
public void notifySubscriber(final Subscriber subscriber, final Event event) {
    LOGGER.debug("[NotifyCenter] the {} will received by {}", event, subscriber);
    //将订阅者对事件的处理构建成一个Runnable
    final Runnable job = () -> subscriber.onEvent(event);
    //获取订阅者处理事件的Executer，这里的设计逻辑是：将异步或者同步的决定权交给订阅者本身，
    //是否异步处理决定于订阅者对executor()方法的实现方式
    final Executor executor = subscriber.executor();
    
    if (executor != null) {
        executor.execute(job);
    } else {
        try {
        	//如果没有executor，那就直接执行Runnable
            job.run();
        } catch (Throwable e) {
            LOGGER.error("Event callback exception: ", e);
        }
    }
}

看了这么多代码，简单对DefaultPublisher做个总结：

• 是什么？
  • 事件通知的默认实现，用于处理事件通知逻辑
  • 继承了线程
• 有什么？
  • 订阅者集合
  • 事件集合
• 做什么？
  • 处理事件通知，同时处理订阅者的相关逻辑（添加/删除）
• 怎么做？
  • 初始化方法中会启动线程
  • run()处理事件，从自己的事件集合中获取事件，然后交给订阅者去处理

class DefaultSharePublisher

这个类继承了DefaultPublisher，设计的目的应该是处理一些比较耗时的事件。本身有一个map做缓存，没有什么特别的逻辑。

private final Map<Class<? extends SlowEvent>, Set<Subscriber>> subMappings = new ConcurrentHashMap<>();

这个类去处理事件的逻辑还是其父类DefaultPublisher的notifySubscriber(subscriber, event)方法。我们上文已经看过这个方法了。

---

回过头再看NotifyCenter

看完EventPublisher之后，别忘记我们为什么出发。
回过头再看NotifyCenter：我们从接收一个事件开始，从头梳理代码。

接收事件

/**
 * Request publisher publish event Publishers load lazily, calling publisher. Start () only when the event is
 * actually published.
 *
 * @param event class Instances of the event.
 */
public static boolean publishEvent(final Event event) {
    try {
        return publishEvent(event.getClass(), event);
    } catch (Throwable ex) {
        LOGGER.error("There was an exception to the message publishing : ", ex);
        return false;
    }
}

发布事件

/**
 * Request publisher publish event Publishers load lazily, calling publisher.
 *
 * @param eventType class Instances type of the event type.
 * @param event     event instance.
 */
private static boolean publishEvent(final Class<? extends Event> eventType, final Event event) {
    if (ClassUtils.isAssignableFrom(SlowEvent.class, eventType)) {
    	//当前事件如果是SlowEvent，那么交与sharePublisher处理
        return INSTANCE.sharePublisher.publish(event);
    }
    //获取Event的类名
    final String topic = ClassUtils.getCanonicalName(eventType);
    
    //根据事件的类名，从map中获取EventPublisher去处理当前事件
    EventPublisher publisher = INSTANCE.publisherMap.get(topic);
    if (publisher != null) {
        return publisher.publish(event);
    }
    if (event.isPluginEvent()) {
        return true;
    }
    LOGGER.warn("There are no [{}] publishers for this event, please register", topic);
    return false;
}

ps：Class.getCanonicalName()的用法

EventPublisher是如何添加到publisherMap中的？

public static void registerSubscriber(final Subscriber consumer) {
    registerSubscriber(consumer, DEFAULT_PUBLISHER_FACTORY);
}
public static void registerSubscriber(final Subscriber consumer, final EventPublisherFactory factory) {
    // If you want to listen to multiple events, you do it separately,
    // based on subclass's subscribeTypes method return list, it can register to publisher.
    //SmartSubscriber是为了处理一个订阅者订阅多个事件而设计的
    if (consumer instanceof SmartSubscriber) {
        for (Class<? extends Event> subscribeType : ((SmartSubscriber) consumer).subscribeTypes()) {
            // For case, producer: defaultSharePublisher -> consumer: smartSubscriber.
            if (ClassUtils.isAssignableFrom(SlowEvent.class, subscribeType)) {
                INSTANCE.sharePublisher.addSubscriber(consumer, subscribeType);
            } else {
                // For case, producer: defaultPublisher -> consumer: subscriber.
                addSubscriber(consumer, subscribeType, factory);
            }
        }
        return;
    }
    
    final Class<? extends Event> subscribeType = consumer.subscribeType();
    if (ClassUtils.isAssignableFrom(SlowEvent.class, subscribeType)) {
        INSTANCE.sharePublisher.addSubscriber(consumer, subscribeType);
        return;
    }
    //添加订阅者
    addSubscriber(consumer, subscribeType, factory);
}

private static void addSubscriber(final Subscriber consumer, Class<? extends Event> subscribeType,
        EventPublisherFactory factory) {
    
    final String topic = ClassUtils.getCanonicalName(subscribeType);
    synchronized (NotifyCenter.class) {
        // MapUtils.computeIfAbsent is a unsafe method.
        MapUtil.computeIfAbsent(INSTANCE.publisherMap, topic, factory, subscribeType, ringBufferSize);
    }
    EventPublisher publisher = INSTANCE.publisherMap.get(topic);
    if (publisher instanceof ShardedEventPublisher) {
        ((ShardedEventPublisher) publisher).addSubscriber(consumer, subscribeType);
    } else {
        publisher.addSubscriber(consumer);
    }
}

最后的一个方法中我们看到了Map publisherMap的数据添加逻辑：根据Event的类名做为key，EventPublisher作为value。
同时我们也看到了事件订阅者是如何和事件发布者建立关系的：publisher.addSubscriber(consumer);，publisher中有一个集合，用于存储事件的订阅者。

总结

好了，总结下NotifyCenter:

• 是什么？
  • 统一的事件处理类，所有的事件通知都借助此类去处理
  • 组织事件-发布者的关联关系
• 有什么？
  • 事件处理类EventPublisher
  • 共享事件处理类DefaultSharePublisher，用于处理耗时的事件
  • Map publisherMap
• 做什么？
  • 注册发布者到publisherMap
  • 接收事件之后交与对应的订阅者处理
• 怎么做？
  • 根据Event的类名去publisherMap 中获取对应的EventPublisher，然后发布事件，交与对应的订阅者处理事件

本篇主要讲了NotifyCenter相关的事件代码逻辑，特别对 EventPublisher做了详细的分析。我们从中看到了一个事件发布与订阅解耦的设计方式，很多框架中都会使用事件驱动去设计发布-订阅模型，日常工作中如果遇到类似的情景，也可以尝试使用这种方式去实现。