RabbitMQ入门_05_多线程消费同一队列

A. 多线程消费同一队列

参考资料：https://www.rabbitmq.com/tutorials/tutorial-two-java.html

消费一条消息往往比产生一条消息慢很多，为了防止消息积压，一般需要开启多个工作线程同时消费消息。在 RabbitMQ 中，我们可以创建多个 Consumer 消费同一队列。示意图如下：

gordon.study.rabbitmq.workqueue.Sender.java

public class Sender {
    private static final String QUEUE_NAME = "tasks";
    private String name;
    public Sender(String name) {
        this.name = name;
    }
    public void work() throws Exception {
        ConnectionFactory factory = new ConnectionFactory();
        factory.setHost("localhost");
        Connection connection = factory.newConnection();
        Channel channel = connection.createChannel();
        channel.queueDeclare(QUEUE_NAME, false, false, false, null);
        for (int i = 0; i < 10;) {
            String message = "NO. " + ++i;
            TimeUnit.MILLISECONDS.sleep(100);
            channel.basicPublish("", QUEUE_NAME, null, message.getBytes("UTF-8"));
            System.out.printf("(%1$s)[===>%2$s    ] %3$s\n", name, ":" + QUEUE_NAME, message);
        }
        channel.close();
        connection.close();
    }
}

gordon.study.rabbitmq.workqueue.Receiver.java

public class Receiver {
    private static final String QUEUE_NAME = "tasks";
    private String name;
    private int sleepTime;
    public Receiver(String name, int sleepTime) {
        this.name = name;
        this.sleepTime = sleepTime;
    }
    public void work() throws Exception {
        ConnectionFactory factory = new ConnectionFactory();
        factory.setHost("localhost");
        Connection connection = factory.newConnection();
        Channel channel = connection.createChannel();
        channel.queueDeclare(QUEUE_NAME, false, false, false, null);
        Consumer consumer = new DefaultConsumer(channel) {
            @Override
            public void handleDelivery(String consumerTag, Envelope envelope, AMQP.BasicProperties properties, byte[] body)
                    throws IOException {
                String message = new String(body, "UTF-8");
                System.out.printf(" [    %2$s<===](%1$s) %3$s\n", name, QUEUE_NAME, message);
                try {
                    TimeUnit.MILLISECONDS.sleep(sleepTime);
                } catch (InterruptedException e) {
                }
            }
        };
        channel.basicConsume(QUEUE_NAME, true, consumer);
    }
}

gordon.study.rabbitmq.workqueue.Test01.java

public class Test01 {
    public static void main(String[] args) throws Exception {
        Receiver recv1 = new Receiver("A", 200);
        recv1.work();
        Receiver recv2 = new Receiver("B", 200);
        recv2.work();
        Sender sender = new Sender("S");
        sender.work();
    }
}

运行 Test01，发现 A、B 两个消费者轮流获取 S 发送的消息。

RabbitMQ 默认将消息顺序发送给下一个消费者，这样，每个消费者会得到相同数量的消息。即，轮询（round-robin）分发消息。

轮询很好，可是如果两个消费者消费能力不一样呢？

gordon.study.rabbitmq.workqueue.Test02SlowConsumer.java

public class Test02SlowConsumer {
    public static void main(String[] args) throws Exception {
        Receiver recv1 = new Receiver("A", 200);
        recv1.work();
        Receiver recv2 = new Receiver("B", 800);
        recv2.work();
        Sender sender = new Sender("S");
        sender.work();
    }
}

将消费者B 的消费时间提高到800毫秒，问题就出现了：B 依然分到了一半消息，需要运行很久才能处理完。

B. 公平分发（fair dispatch）

怎样才能做到按照每个消费者的能力分配消息呢？联合使用 Qos 和 Acknowledge 就可以做到。

gordon.study.rabbitmq.workqueue.QosAcknowledgeReceiver.java

public class QosAcknowledgeReceiver {
    private static final String QUEUE_NAME = "tasks";
    private String name;
    private int sleepTime;
    public QosAcknowledgeReceiver(String name, int sleepTime) {
        this.name = name;
        this.sleepTime = sleepTime;
    }
    public void work() throws Exception {
        ConnectionFactory factory = new ConnectionFactory();
        factory.setHost("localhost");
        Connection connection = factory.newConnection();
        final Channel channel = connection.createChannel();
        channel.queueDeclare(QUEUE_NAME, false, false, false, null);
        channel.basicQos(1);
        Consumer consumer = new DefaultConsumer(channel) {
            @Override
            public void handleDelivery(String consumerTag, Envelope envelope, AMQP.BasicProperties properties, byte[] body)
                    throws IOException {
                String message = new String(body, "UTF-8");
                System.out.printf(" [    %2$s<===](%1$s) %3$s\n", name, QUEUE_NAME, message);
                try {
                    TimeUnit.MILLISECONDS.sleep(sleepTime);
                } catch (InterruptedException e) {
                }
                channel.basicAck(envelope.getDeliveryTag(), false);
            }
        };
        channel.basicConsume(QUEUE_NAME, false, consumer);
    }
}

代码第22行的 basicQos 方法设置了当前信道最大预获取（prefetch）消息数量为1。消息从队列异步推送给消费者，消费者的 ack 也是异步发送给队列，从队列的视角去看，总是会有一批消息已推送但尚未获得 ack 确认，Qos 的 prefetchCount 参数就是用来限制这批未确认消息数量的。设为1时，队列只有在收到消费者发回的上一条消息 ack 确认后，才会向该消费者发送下一条消息。prefetchCount 的默认值为0，即没有限制，队列会将所有消息尽快发给消费者。

查看 basicQos 重载方法，发现几个有趣的特性（参考https://www.rabbitmq.com/consumer-prefetch.html）：

• basicQos 中 prefetchSize 参数通过消息的总字节数来限制队列推送消息的速度
• prefetchSize 与 prefetchCount 可以同时设置，达到任何一个限制，则队列暂停推送消息
• global 参数表示前两个参数的作用域，true 表示限制是针对信道的，false 表示限制是针对消费者的（我还没试过一个信道支持多个消费者的例子，样例代码见下方）
• 可以对同一个信道同时设置 global 为 true 和 false 的 Qos，表示队列要考虑每个消费者的限制，同时还要考虑整个信道的限制
• 看起来API注释是错的，因为 global 默认是 false，所以第22行代码应该是把当前信道上每个消费者（当然，上面的例子中只有一个）的 prefetchCount 设为 1

Channel channel = ...;
Consumer consumer1 = ...;
Consumer consumer2 = ...;
channel.basicQos(10, false); // Per consumer limit
channel.basicQos(15, true);  // Per channel limit
channel.basicConsume("my-queue1", false, consumer1);
channel.basicConsume("my-queue2", false, consumer2);

第37行代码将 autoAck 设为 false，向 Broker 发送 ack 响应的任务就交给开发人员了。

第34行代码在任务真正完成后，调用 basicAck 方法主动通知队列消息已成功消费。当队列收到 ack 确认后，会把下一条消息推送过来，并将该消息从队列中删除。

Qos 方案示意图如下：

gordon.study.rabbitmq.workqueue.Test03FairDispatch.java

public class Test03FairDispatch {
    public static void main(String[] args) throws Exception {
        QosAcknowledgeReceiver recv1 = new QosAcknowledgeReceiver("A", 200);
        recv1.work();
        QosAcknowledgeReceiver recv2 = new QosAcknowledgeReceiver("B", 800);
        recv2.work();
        Sender sender = new Sender("S");
        sender.work();
    }
}

运行Test03，可以看到 RabbitMQ 按照消费者的实际能力分配消息。