本篇随笔将汇总一些我对消息队列 RabbitMQ 的认识,顺便谈谈其在高并发和秒杀系统中的具体应用。
想了下,还是先抛出一个简单示例,随后再根据其具体应用场景进行扩展,我觉得这样表述条理更清晰些。
RabbitConfig:
@Configuration
public class RabbitConfig {
@Bean
public Queue callQueue() {
return new Queue(MQConstant.CALL);
}
}
Client:
@Component
public class Client {
@Autowired
private RabbitTemplate rabbitTemplate;
public void sendCall(String content) {
for (int i = 0; i < 10000; i++) {
String message = i + "-" + content;
System.out.println(String.format("Sender: %s", message));
rabbitTemplate.convertAndSend(MQConstant.CALL, message);
}
}
}
Server:
@Component
public class Server {
@RabbitHandler
@RabbitListener(queues = MQConstant.CALL)
public void callProcess(String message) throws InterruptedException {
Thread.sleep(1000);
System.out.println(String.format("Receiver: reply(\"%s\") Yes, I just saw your message!", message));
}
}
Result:
Sender: Hello, are you there!
Receiver: reply("Hello, are you there!") Yes, I just saw your message!
以上示例会在 rabbitmq 中创建一条队列 CALL, 消息在其中等待消费:
在此基础上的简单扩展我就不再写案例了,比如领域模块完成了其核心业务规则之后可能需要更新缓存、写个邮件、记个复杂日志、做个统计报表等等,这些不需要及时反馈或者耗时的附属业务都可以通过异步队列分发,以此来提升核心业务的响应速度,同时如此处理能让领域边界更加清晰,代码的可维护性和持续拓展的能力也会有所提升。
上个示例中我提到的应用场景是解耦和通知,再接着扩展,因其具备良好的缓冲性质,所以还有一个非常适合的应用场景那就是削峰。对于突如其来的极高并发请求,我们可以先瞬速地将其加入队列并回复用户一个友好提示,然后服务器可在其能承受的范围内慢慢处理,以此来防止突发的 CPU 和内存 “爆表”。
改造之后对于发送方来说当然是比较爽的,他只是将请求加入消息队列而已,处理压力都归到了消费端。接着思考,这样处理有没有副作用?如果这个请求刚好是线程阻塞的,那还要加入队列慢慢排队处理,那不是完蛋了,用户要猴年马月才能得到反馈?所以针对此,我觉得应该将消费端的方法改为异步调用(即多线程)以提升吞吐量,在 Spring Boot 中的写法也非常简单:
@Component
public class Server {
@Async
@RabbitHandler
@RabbitListener(queues = MQConstant.CALL)
public void callProcess(String message) throws InterruptedException {
Thread.sleep(100);
System.out.println(String.format("Receiver: reply(\"%s\") Yes, I just saw your message!", message));
}
}
参照示例一的方法,我发布了 10000 条消息加入队列,且消费端的调用每次阻塞一秒,那可有意思了,什么时候能处理完?但如果开几百个线程同时处理的话,那几十秒就够了,当然具体多少合适还应根据具体的业务场景和服务器配置酌情考虑。另外,别忘了配线程池:
@Configuration
public class AsyncConfig {
@Bean
public Executor asyncExecutor(){
ThreadPoolTaskExecutor executor = new ThreadPoolTaskExecutor();
executor.setCorePoolSize(10);
executor.setMaxPoolSize(500);
executor.setQueueCapacity(10);
executor.setThreadNamePrefix("MyExecutor-");
executor.setRejectedExecutionHandler(new ThreadPoolExecutor.CallerRunsPolicy());
executor.initialize();
return executor;
}
}
RabbitMQ 可能为 N 个应用同时提供服务,要是你和你的蓝颜知己突然心有灵犀,在不同的业务上使用了同一个 routingKey,想想就刺激。因此,队列多了自然要进行分组管理,限定好 Exchange 的规则,接下来就可以独自玩耍了。
MQConstant:
public class MQConstant {
public static final String EXCHANGE = "YOUCLK-MESSAGE-EXCHANGE";
public static final String CALL = MQConstant.EXCHANGE + ".CALL";
public static final String ALL = MQConstant.EXCHANGE + ".#";
}
RabbitConfig:
@Configuration
public class RabbitConfig {
@Bean
public Queue callQueue() {
return new Queue(MQConstant.CALL);
}
@Bean
TopicExchange exchange() {
return new TopicExchange(MQConstant.EXCHANGE);
}
@Bean
Binding bindingExchangeMessage(Queue queueMessage, TopicExchange exchange) {
return BindingBuilder.bind(queueMessage).to(exchange).with(MQConstant.ALL);
}
}
此时我们再去查队列 CALL,可以看到已经绑定了Exchange:
当然 Exchange 的作用远不止如此,以上示例为 Topic 模式,除此之外还有 Direct、Headers 和 Fanout 模式,写法都差不多,感兴趣的童鞋可以去查看 “官方文档” 进行更深入了解。
延时任务的场景相信小伙伴们都接触过,特别是抢购的时候,在规定时间内未付款订单就被回收了。微信支付的 API 里面也有一个支付完成后的延时再确认消息推送,实现原理应该都差不多。
利用 RabbitMQ 实现该功能首先要了解他的两个特性,分别是 Time-To-Live Extensions 和 Dead Letter Exchanges,字面意思上就能理解个大概,一个是生存时间,一个是死信。整个过程也很容易理解,TTL 相当于一个缓冲队列,等待其过期之后消息会由 DLX 转发到实际消费队列,如此便实现了他的延时过程。
MQConstant:
public class MQConstant {
public static final String PER_DELAY_EXCHANGE = "PER_DELAY_EXCHANGE";
public static final String DELAY_EXCHANGE = "DELAY_EXCHANGE";
public static final String DELAY_CALL_TTL = "DELAY_CALL_TTL";
public static final String CALL = "CALL";
}
ExpirationMessagePostProcessor:
public class ExpirationMessagePostProcessor implements MessagePostProcessor {
private final Long ttl;
public ExpirationMessagePostProcessor(Long ttl) {
this.ttl = ttl;
}
@Override
public Message postProcessMessage(Message message) throws AmqpException {
message.getMessageProperties()
.setExpiration(ttl.toString());
return message;
}
}
Client:
@Component
public class Client {
@Autowired
private RabbitTemplate rabbitTemplate;
public void sendCall(String content) {
for (int i = 1; i <= 3; i++) {
long expiration = i * 5000;
String message = i + "-" + content;
System.out.println(String.format("Sender: %s", message));
rabbitTemplate.convertAndSend(MQConstant.DELAY_CALL_TTL, (Object) message, new ExpirationMessagePostProcessor(expiration));
}
}
}
Server:
@Component
public class Server {
@Async
@RabbitHandler
@RabbitListener(queues = MQConstant.CALL)
public void callProcess(String message) throws InterruptedException {
String date = (new SimpleDateFormat("HH:mm:ss")).format(new Date());
System.out.println(String.format("Receiver: reply(\"%s\") Yes, I just saw your message!- %s", message, date));
}
}
Result:
Sender: 1-Hello, are you there!
Sender: 2-Hello, are you there!
Sender: 3-Hello, are you there!
Receiver: reply("1-Hello, are you there!") Yes, I just saw your message!- 23:04:12
Receiver: reply("2-Hello, are you there!") Yes, I just saw your message!- 23:04:17
Receiver: reply("3-Hello, are you there!") Yes, I just saw your message!- 23:04:22
结果一目了然,分别在队列中延迟了 5秒,10秒,15秒,当然,以上只是我的简单示例,童鞋们可翻阅官方文档(“ ttl ” && “ dlx ”)进一步深入学习。