【RocketMQ】消息拉模式分析

时间:2023-01-20 11:06:20

RocketMQ有两种获取消息的方式,分别为推模式和拉模式。

推模式
推模式在【RocketMQ】消息的拉取一文中已经讲过,虽然从名字上看起来是消息到达Broker后推送给消费者,实际上还是需要消费向Broker发送拉取请求获取消息内容,推模式对应的消息消费实现类为DefaultMQPushConsumerImpl,回顾一下推模式下的消息消费过程:

  1. 消费者在启动的时候做一些初始化工作,它会创建MQClientInstance并进行启动;
  2. MQClientInstance中引用了消息拉取服务PullMessageService和负载均衡服务RebalanceService,它们都继承了ServiceThread,MQClientInstance在启动后也会对它们进行启动,所以消息拉取线程和负载均衡线程也就启动了;
  3. 负载均衡服务启动后,会对该消费者订阅的主题进行负载均衡,为消费者分配消息队列,并创建PullRequest拉取请求,用于拉取消息;
  4. PullMessageService中等待阻塞队列中PullRequest拉取请求的到来,接着会调用DefaultMQPushConsumerImplpullMessage方法进行消息拉取;
  5. 消费者向Broker发送拉取消息的请求,从Broker拉取消息;
  6. 消费者对Broker返回的响应数据进行处理,解析消息进行消费;

推模式下进行消息消费的例子:

@RunWith(MockitoJUnitRunner.class)
public class DefaultMQPushConsumerTest {
    private String consumerGroup;
    private String topic = "FooBar";
    private String brokerName = "BrokerA";
    private MQClientInstance mQClientFactory;

    @Mock
    private MQClientAPIImpl mQClientAPIImpl;
    private static DefaultMQPushConsumer pushConsumer;

    @Before
    public void init() throws Exception {
        // ...
        // 消费者组
        consumerGroup = "FooBarGroup" + System.currentTimeMillis();
        // 实例化DefaultMQPushConsumer
        pushConsumer = new DefaultMQPushConsumer(consumerGroup);
        pushConsumer.setNamesrvAddr("127.0.0.1:9876");
        // 设置拉取间隔
        pushConsumer.setPullInterval(60 * 1000);
        // 注册消息监听器
        pushConsumer.registerMessageListener(new MessageListenerConcurrently() {
            @Override
            public ConsumeConcurrentlyStatus consumeMessage(List<MessageExt> msgs,
                ConsumeConcurrentlyContext context) {
                Optional.ofNullable(result).orElse(new ArrayList<MessageExt>()).stream().forEach(x-> {
                    // 处理消息
                    System.out.println(new String(x.getBody()));
                });
                return null;
            }
        });
        // ...
        // 设置订阅的主题
        pushConsumer.subscribe(topic, "*");
        // 启动消费者
        pushConsumer.start();
    }
}

消息推模式的详细过程可参考【RocketMQ】消息的拉取,接下来我们看一下拉模式。

拉模式
首先来看一下拉模式下进行消息消费的例子,拉模式下需要消费者不断调用poll方法获取消息,底层是一个阻塞队列,如果队列中没有数据,会进入等待直到队列中增加了数据:

 private void testPull() {
        // 创建DefaultLitePullConsumer
        DefaultLitePullConsumer litePullConsumer = new DefaultLitePullConsumer("LitePullConsumerGroup");;
        try {
            litePullConsumer.setNamesrvAddr("127.0.0.1:9876");
            litePullConsumer.subscribe("LitePullConsumerTest", "*");
            litePullConsumer.start();
            litePullConsumer.setPollTimeoutMillis(20 * 1000);
            while(true) {
                // 获取消息
                List<MessageExt> result = litePullConsumer.poll();
                Optional.ofNullable(result).orElse(new ArrayList<MessageExt>()).stream().forEach(x-> {
                    // 处理消息
                    System.out.println(new String(x.getBody()));
                });
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            litePullConsumer.shutdown();
        }
    }

推模式与拉模式的区别
对比上面推模式进行消费的例子,从使用方式上来讲,推模式不需要消费者主动去拉取消息,只需要注册消息监听器,当有消息到达时,触发consumeMessage方法进行消息消费,从表面上看就像是Broker主动推送给消费者一样,所以叫做推模式,尽管底层还是需要消费者发起拉取请求向Broker拉取消息

拉模式在使用方式上,需要消费者主动调用poll方法获取消息,从表面上看消费者需要不断主动进行消息拉取,所以叫做拉模式。

拉模式实现原理

拉模式下对应的消息拉取实现类为DefaultLitePullConsumerImpl,在DefaultLitePullConsumerDefaultMQPullConsumer被标注了@Deprecated,已不推荐使用)的构造函数中,可以看到对其进行了实例化,并在start方进行了启动:

public class DefaultLitePullConsumer extends ClientConfig implements LitePullConsumer {
    // 拉模式下默认的消息拉取实现类
    private final DefaultLitePullConsumerImpl defaultLitePullConsumerImpl;

    public DefaultLitePullConsumer(final String namespace, final String consumerGroup, RPCHook rpcHook) {
        this.namespace = namespace;
        this.consumerGroup = consumerGroup;
        // 创建DefaultLitePullConsumerImpl
        defaultLitePullConsumerImpl = new DefaultLitePullConsumerImpl(this, rpcHook);
    }

    @Override
    public void start() throws MQClientException {
        setTraceDispatcher();
        setConsumerGroup(NamespaceUtil.wrapNamespace(this.getNamespace(), this.consumerGroup));
        // 启动DefaultLitePullConsumerImpl
        this.defaultLitePullConsumerImpl.start();
        // ...
    }
}

与消息推模式类似,DefaultLitePullConsumerImpl的start的方法主要做一些初始化的工作:

  1. 初始化客户端实例对象mQClientFactory,对应实现类为MQClientInstance,拉取服务线程、负载均衡线程都是通过MQClientInstance启动的;
  2. 初始化负载均衡类,拉模式对应的负载均衡类为RebalanceLitePullImpl
  3. 创建消息拉取API对象PullAPIWrapper,用于向Broker发送拉取消息的请求;
  4. 初始化消息拉取偏移量;
  5. 启动一些定时任务;
public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    public synchronized void start() throws MQClientException {
        switch (this.serviceState) {
            case CREATE_JUST:
                this.serviceState = ServiceState.START_FAILED;
                this.checkConfig();
                if (this.defaultLitePullConsumer.getMessageModel() == MessageModel.CLUSTERING) {
                    this.defaultLitePullConsumer.changeInstanceNameToPID();
                }
                // 初始化MQClientInstance
                initMQClientFactory();
                // 初始化负载均衡
                initRebalanceImpl();
                // 初始化消息拉取API对象
                initPullAPIWrapper();
                // 初始化拉取偏移量
                initOffsetStore();
                // 启动MQClientInstance
                mQClientFactory.start();
                // 启动一些定时任务
                startScheduleTask();
                this.serviceState = ServiceState.RUNNING;
                log.info("the consumer [{}] start OK", this.defaultLitePullConsumer.getConsumerGroup());
                operateAfterRunning();
                break;
            case RUNNING:
            case START_FAILED:
            case SHUTDOWN_ALREADY:
                throw new MQClientException("The PullConsumer service state not OK, maybe started once, "
                    + this.serviceState
                    + FAQUrl.suggestTodo(FAQUrl.CLIENT_SERVICE_NOT_OK),
                    null);
            default:
                break;
        }
    }
}

负载均衡

拉取模式对应的负载均衡类为RebalanceLitePullImpl(推模式使用的是RebalanceService),在initRebalanceImpl方法中设置了消费者组、消费模式、分配策略等信息:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    
    // 实例化,拉模式使用的是RebalanceLitePullImpl
    private RebalanceImpl rebalanceImpl = new RebalanceLitePullImpl(this);

    private void initRebalanceImpl() {
        // 设置消费者组
        this.rebalanceImpl.setConsumerGroup(this.defaultLitePullConsumer.getConsumerGroup());
        // 设置消费模式
        this.rebalanceImpl.setMessageModel(this.defaultLitePullConsumer.getMessageModel());
        // 设置分配策略
        this.rebalanceImpl.setAllocateMessageQueueStrategy(this.defaultLitePullConsumer.getAllocateMessageQueueStrategy());
        // 设置mQClientFactory
        this.rebalanceImpl.setmQClientFactory(this.mQClientFactory);
    }
}

【RocketMQ】消息的拉取一文中已经讲到过,消费者启动后会进行负载均衡,对每个主题进行负载均衡,拉模式下处理逻辑也是如此,所以这里跳过中间的过程,进入到rebalanceByTopic方法,可以负载均衡之后如果消费者负载的ProcessQueue发生了变化,会调用messageQueueChanged方法触发变更事件:

public abstract class RebalanceImpl {
     private void rebalanceByTopic(final String topic, final boolean isOrder) {
        switch (messageModel) {
            case BROADCASTING: {
                // ...
            }
            case CLUSTERING: {
                Set<MessageQueue> mqSet = this.topicSubscribeInfoTable.get(topic);
                List<String> cidAll = this.mQClientFactory.findConsumerIdList(topic, consumerGroup);
                // ...
                if (mqSet != null && cidAll != null) {
                    // ...
                    try {
                        // 分配消息队列
                        allocateResult = strategy.allocate(this.consumerGroup, this.mQClientFactory.getClientId(), mqAll, cidAll);
                    } catch (Throwable e) {
                        log.error("AllocateMessageQueueStrategy.allocate Exception. allocateMessageQueueStrategyName={}", strategy.getName(),
                            e);
                        return;
                    }

                    Set<MessageQueue> allocateResultSet = new HashSet<MessageQueue>();
                    if (allocateResult != null) {
                        allocateResultSet.addAll(allocateResult);
                    }
                    // 更新处理队列
                    boolean changed = this.updateProcessQueueTableInRebalance(topic, allocateResultSet, isOrder);
                    if (changed) {
                        log.info(
                            "rebalanced result changed. allocateMessageQueueStrategyName={}, group={}, topic={}, clientId={}, mqAllSize={}, cidAllSize={}, rebalanceResultSize={}, rebalanceResultSet={}",
                            strategy.getName(), consumerGroup, topic, this.mQClientFactory.getClientId(), mqSet.size(), cidAll.size(),
                            allocateResultSet.size(), allocateResultSet);
                        // 触发变更事件
                        this.messageQueueChanged(topic, mqSet, allocateResultSet);
                    }
                }
                break;
            }
            default:
                break;
        }
    }
}

触发消息队列变更事件

RebalanceLitePullImplmessageQueueChanged方法中又调用了MessageQueueListenermessageQueueChanged方法触发消息队列改变事件:

public class RebalanceLitePullImpl extends RebalanceImpl {
    @Override
    public void messageQueueChanged(String topic, Set<MessageQueue> mqAll, Set<MessageQueue> mqDivided) {
        MessageQueueListener messageQueueListener = this.litePullConsumerImpl.getDefaultLitePullConsumer().getMessageQueueListener();
        if (messageQueueListener != null) {
            try {
                // 触发改变事件
                messageQueueListener.messageQueueChanged(topic, mqAll, mqDivided);
            } catch (Throwable e) {
                log.error("messageQueueChanged exception", e);
            }
        }
    }
}

MessageQueueListenerImplDefaultLitePullConsumerImpl的内部类,在messageQueueChanged方法中,不管是广播模式还是集群模式,都会调用updatePullTask更新拉取任务:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    class MessageQueueListenerImpl implements MessageQueueListener {
        @Override
        public void messageQueueChanged(String topic, Set<MessageQueue> mqAll, Set<MessageQueue> mqDivided) {
            MessageModel messageModel = defaultLitePullConsumer.getMessageModel();
            switch (messageModel) {
                case BROADCASTING:
                    updateAssignedMessageQueue(topic, mqAll);
                    updatePullTask(topic, mqAll); // 更新拉取任务
                    break;
                case CLUSTERING:
                    updateAssignedMessageQueue(topic, mqDivided);
                    updatePullTask(topic, mqDivided); // 更新拉取任务
                    break;
                default:
                    break;
            }
        }
    }
}

更新拉取任务

在updatePullTask方法中,从拉取任务表taskTable中取出了所有的拉取任务进行遍历,taskTable中记录了之前分配的拉取任务,负载均衡之后可能发生变化,所以需要对其进行更新,这一步主要是处理原先分配给当前消费者的消息队列,在负载均衡之后不再由当前消费者负责,所以需要从taskTable中删除,之后调用startPullTask启动拉取任务:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    private final ConcurrentMap<MessageQueue, PullTaskImpl> taskTable =
        new ConcurrentHashMap<MessageQueue, PullTaskImpl>();

    private void updatePullTask(String topic, Set<MessageQueue> mqNewSet) {
        // 从拉取任务表中获取之前分配的消息队列进行遍历
        Iterator<Map.Entry<MessageQueue, PullTaskImpl>> it = this.taskTable.entrySet().iterator();
        while (it.hasNext()) {
            Map.Entry<MessageQueue, PullTaskImpl> next = it.next();
            // 如果与重新进行负载均衡的主题一致
            if (next.getKey().getTopic().equals(topic)) {
                // 如果重新分配的消息队列集合中不包含此消息独立
                if (!mqNewSet.contains(next.getKey())) {
                    next.getValue().setCancelled(true);
                    // 从任务表移除
                    it.remove();
                }
            }
        }
        // 启动拉取任务
        startPullTask(mqNewSet);
    }
}

提交拉取任务

startPullTask方法入参中传入的是负载均衡后重新分配的消息队列集合,在startPullTask中会对重新分配的集合进行遍历,如果taskTable中不包含某个消息队列,就构建PullTaskImpl对象,加入taskTable,这一步主要是处理负载均衡后新增的消息队列,为其构建PullTaskImpl加入到taskTable,之后将拉取消息的任务PullTaskImpl提交到线程池周期性的执行:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {

    private void startPullTask(Collection<MessageQueue> mqSet) {
        // 遍历最新分配的消息队列集合
        for (MessageQueue messageQueue : mqSet) {
            // 如果任务表中不包含
            if (!this.taskTable.containsKey(messageQueue)) {
                // 创建拉取任务
                PullTaskImpl pullTask = new PullTaskImpl(messageQueue);
                // 加入到任务表
                this.taskTable.put(messageQueue, pullTask);
                // 将任务提交到线程池定时执行
                this.scheduledThreadPoolExecutor.schedule(pullTask, 0, TimeUnit.MILLISECONDS);
            }
        }
    }
}

拉取消息

PullTaskImpl继承了Runnable,在run方法中的处理逻辑如下:

  1. 获取消息队列对应处理队列ProcessQueue;
  2. 获取消息拉取偏移量,也就是从何处开始拉取消息;
  3. 调用pull方法进行消息拉取;
  4. 判断拉取结果,如果拉取到了消息,将拉取到的结果封装为ConsumeRequest进行提交,也就是放到了阻塞队列中,后续消费者从队列中获取数据进行消费;
   public class PullTaskImpl implements Runnable {
        private final MessageQueue messageQueue;
        private volatile boolean cancelled = false;
        private Thread currentThread;

        @Override
        public void run() {
            // 如果未取消
            if (!this.isCancelled()) {
                this.currentThread = Thread.currentThread();
                // ...
                // 获取消息队列对应的ProcessQueue
                ProcessQueue processQueue = assignedMessageQueue.getProcessQueue(messageQueue);
                // ...  跳过一系列校验
                long offset = 0L;
                try {
                    // 获取拉取偏移量
                    offset = nextPullOffset(messageQueue);
                } catch (Exception e) {
                    log.error("Failed to get next pull offset", e);
                    scheduledThreadPoolExecutor.schedule(this, PULL_TIME_DELAY_MILLS_ON_EXCEPTION, TimeUnit.MILLISECONDS);
                    return;
                }

                if (this.isCancelled() || processQueue.isDropped()) {
                    return;
                }
                long pullDelayTimeMills = 0;
                try {
                    SubscriptionData subscriptionData;
                    // 获取主题
                    String topic = this.messageQueue.getTopic();
                    // 获取主题对应的订阅信息SubscriptionData
                    if (subscriptionType == SubscriptionType.SUBSCRIBE) {
                        subscriptionData = rebalanceImpl.getSubscriptionInner().get(topic);
                    } else {
                        subscriptionData = FilterAPI.buildSubscriptionData(topic, SubscriptionData.SUB_ALL);
                    }
                    // 拉取消息
                    PullResult pullResult = pull(messageQueue, subscriptionData, offset, defaultLitePullConsumer.getPullBatchSize());
                    if (this.isCancelled() || processQueue.isDropped()) {
                        return;
                    }
                    // 判断拉取结果
                    switch (pullResult.getPullStatus()) {
                        case FOUND: // 如果获取到了数据
                            final Object objLock = messageQueueLock.fetchLockObject(messageQueue);
                            synchronized (objLock) { // 加锁
                                if (pullResult.getMsgFoundList() != null && !pullResult.getMsgFoundList().isEmpty() && assignedMessageQueue.getSeekOffset(messageQueue) == -1) {
                                    processQueue.putMessage(pullResult.getMsgFoundList());
                                    // 将拉取结果封装为ConsumeRequest,提交消费请求
                                    submitConsumeRequest(new ConsumeRequest(pullResult.getMsgFoundList(), messageQueue, processQueue));
                                }
                            }
                            break;
                        case OFFSET_ILLEGAL:
                            log.warn("The pull request offset illegal, {}", pullResult.toString());
                            break;
                        default:
                            break;
                    }
                    updatePullOffset(messageQueue, pullResult.getNextBeginOffset(), processQueue);
                } catch (InterruptedException interruptedException) {
                    log.warn("Polling thread was interrupted.", interruptedException);
                } catch (Throwable e) {
                    pullDelayTimeMills = pullTimeDelayMillsWhenException;
                    log.error("An error occurred in pull message process.", e);
                }
                // ...
            }
        }
    }

submitConsumeRequest方法中可以看到将创建的ConsumeRequest对象放入了阻塞队列consumeRequestCache中:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    // 阻塞队列
    private final BlockingQueue<ConsumeRequest> consumeRequestCache = new LinkedBlockingQueue<ConsumeRequest>();

    private void submitConsumeRequest(ConsumeRequest consumeRequest) {
        try {
            // 放入阻塞队列consumeRequestCache中
            consumeRequestCache.put(consumeRequest);
        } catch (InterruptedException e) {
            log.error("Submit consumeRequest error", e);
        }
    }
}

消息消费

在前面的例子中,可以看到消费者是调用poll方法获取数据的,进入到poll方法中,可以看到是从consumeRequestCache中获取消费请求的,然后从中解析出消息内容返回:

public class DefaultLitePullConsumerImpl implements MQConsumerInner {
    
    public synchronized List<MessageExt> poll(long timeout) {
        try {
            // ...
            long endTime = System.currentTimeMillis() + timeout;
            // 从consumeRequestCache中获取数据进行处理
            ConsumeRequest consumeRequest = consumeRequestCache.poll(endTime - System.currentTimeMillis(), TimeUnit.MILLISECONDS);
            // ...
            if (consumeRequest != null && !consumeRequest.getProcessQueue().isDropped()) {
                // 获取消息内容
                List<MessageExt> messages = consumeRequest.getMessageExts();
                long offset = consumeRequest.getProcessQueue().removeMessage(messages);
                assignedMessageQueue.updateConsumeOffset(consumeRequest.getMessageQueue(), offset);
                this.resetTopic(messages);
                // 返回消息内容
                return messages;
            }
        } catch (InterruptedException ignore) {
        }
        return Collections.emptyList();
    }
}

参考

RocketMQ源码分析之pull模式consumer

RocketMQ版本:4.9.3