20180927 更新
升级 retryCache的容器,修改为 ConcurrentSkipListMap
1 retry的时候按先后顺序尝试
2 hashMap无法自动缩容,在rabbitmq出现问题时,map造成积压,等问题恢复后,map的多余空间无法自动释放,而SkipListMap可以完美避开这个问题
3 在大量插入删除时,SkipList的效率更高
20180710 更新
1 升级spring-rabbit版本,升级到最新版本
2 去除对QueueConsumer的使用,改为使用basicGet方法(消费效率和原来的方式对比,有微弱提升)
3 改进一些打log的细节
20180517 更新
1 retryCache重构,解决rabbitmq挂掉时消息积压的问题
2 部分细节改进
20171120 更新
1 改进一些细节:遍历map时基于entry,增加一定的效率
20170510 更新
1 增加线程池consumer优雅退出机制Runtime.getRuntime().addShutdownHook
2 修改部分log输出方式,将原来的 log.info("exceptin:" + e) 修复为 log.info("exception: ", e)
20161227 更新
1 bug fix: 将messageProcess包裹在try,catch中,避免队列中出现unack的死信息
2 bug分析见http://www.jianshu.com/p/a7edc3322b44
20161205 更新
1 增加topic模式
2 原有的使用direct方式无需更改,本次为兼容性升级,增加了buildTopicMessageSender和buildTopicMessageConsumer方法
3 ThreadPoolConsumer默认为direct方式,可以通过setType("topic")修改为topic模式
20160907 更新
1 解决因网络抖动而引起的发送数据丢失
2 增加retry模块
3 在本地缓存已发送数据,根据ack的确认将已ack的删除
4 定时触发重发未收到ack的数据
5 保证在网络抖动的情况下数据不丢失,但可能会造成数据的重复发送(建议在consumer端做到message处理的幂等性)
最近的一个计费项目,在rpc调用和流式处理之间徘徊了许久,后来选择流式处理。一是可以增加吞吐量,二是事务的控制相比于rpc要容易很多。 确定了流式处理的方式,后续是技术的选型。刚开始倾向于用storm,无奈文档实在太少,折腾起来着实费劲。最终放弃,改用消息队列+微服务的方式实现。
消息队列的选型上,有activemq,rabbitmq,kafka等。最开始倾向于用activemq,因为以前的项目用过,很多代码都是可直接复用的。后来看了不少文章对比,发现rabbitmq对多语言的支持更好一点,同时相比于kafka,牺牲了部分的性能换取了更好的稳定性安全性以及持久化。 最终决定使用rabbitmq。
rabbitmq的官网如下:
对rabbitmq的封装,有几个目标: 1 提供send接口 2 提供consume接口 3 保证消息的事务性处理
所谓事务性处理,是指对一个消息的处理必须严格可控,必须满足原子性,只有两种可能的处理结果: (1) 处理成功,从队列中删除消息 (2) 处理失败(网络问题,程序问题,服务挂了),将消息重新放回队列 为了做到这点,我们使用rabbitmq的手动ack模式,这个后面细说。
1 send接口
public interface MessageSender {
DetailRes send(Object message);
}
send接口相对简单,我们使用spring的RabbitTemplate来实现,代码如下:
//1 构造template, exchange, routingkey等
//2 设置message序列化方法
//3 设置发送确认
//4 构造sender方法
public MessageSender buildMessageSender(final String exchange, final String routingKey, final String queue) throws IOException, TimeoutException {
Connection connection = connectionFactory.createConnection();
//1
buildQueue(exchange, routingKey, queue, connection);
final RabbitTemplate rabbitTemplate = new RabbitTemplate(connectionFactory);
rabbitTemplate.setMandatory(true);
rabbitTemplate.setExchange(exchange);
rabbitTemplate.setRoutingKey(routingKey);
//2
rabbitTemplate.setMessageConverter(new Jackson2JsonMessageConverter());
//3
rabbitTemplate.setConfirmCallback(new RabbitTemplate.ConfirmCallback() {
@Override
public void confirm(CorrelationData correlationData, boolean ack, String cause) {
if (!ack) {
log.info("send message failed: " + cause); //+ correlationData.toString());
throw new RuntimeException("send error " + cause);
}
}
});
//4
return new MessageSender() {
@Override
public DetailRes send(Object message) {
try {
rabbitTemplate.convertAndSend(message);
} catch (RuntimeException e) {
e.printStackTrace();
log.info("send failed " + e);
try {
//retry
rabbitTemplate.convertAndSend(message);
} catch (RuntimeException error) {
error.printStackTrace();
log.info("send failed again " + error);
return new DetailRes(false, error.toString());
}
}
return new DetailRes(true, "");
}
};
}
2 consume接口
public interface MessageConsumer {
DetailRes consume();
}
在consume接口中,会调用用户自己的MessageProcess,接口定义如下:
public interface MessageProcess<T> {
DetailRes process(T message);
}
consume的实现相对来说复杂一点,代码如下:
//1 创建连接和channel
//2 设置message序列化方法
//3 构造consumer
public <T> MessageConsumer buildMessageConsumer(String exchange, String routingKey,
final String queue, final MessageProcess<T> messageProcess) throws IOException {
final Connection connection = connectionFactory.createConnection();
//1
buildQueue(exchange, routingKey, queue, connection);
//2
final MessagePropertiesConverter messagePropertiesConverter = new DefaultMessagePropertiesConverter();
final MessageConverter messageConverter = new Jackson2JsonMessageConverter();
//3
return new MessageConsumer() {
QueueingConsumer consumer;
{
consumer = buildQueueConsumer(connection, queue);
}
@Override
//1 通过delivery获取原始数据
//2 将原始数据转换为特定类型的包
//3 处理数据
//4 手动发送ack确认
public DetailRes consume() {
QueueingConsumer.Delivery delivery = null;
Channel channel = consumer.getChannel();
try {
//1
delivery = consumer.nextDelivery();
Message message = new Message(delivery.getBody(),
messagePropertiesConverter.toMessageProperties(delivery.getProperties(), delivery.getEnvelope(), "UTF-8"));
//2
@SuppressWarnings("unchecked")
T messageBean = (T) messageConverter.fromMessage(message);
//3
DetailRes detailRes = messageProcess.process(messageBean);
//4
if (detailRes.isSuccess()) {
channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false);
} else {
log.info("send message failed: " + detailRes.getErrMsg());
channel.basicNack(delivery.getEnvelope().getDeliveryTag(), false, true);
}
return detailRes;
} catch (InterruptedException e) {
e.printStackTrace();
return new DetailRes(false, "interrupted exception " + e.toString());
} catch (IOException e) {
e.printStackTrace();
retry(delivery, channel);
log.info("io exception : " + e);
return new DetailRes(false, "io exception " + e.toString());
} catch (ShutdownSignalException e) {
e.printStackTrace();
try {
channel.close();
} catch (IOException io) {
io.printStackTrace();
} catch (TimeoutException timeout) {
timeout.printStackTrace();
}
consumer = buildQueueConsumer(connection, queue);
return new DetailRes(false, "shutdown exception " + e.toString());
} catch (Exception e) {
e.printStackTrace();
log.info("exception : " + e);
retry(delivery, channel);
return new DetailRes(false, "exception " + e.toString());
}
}
};
}
3 保证消息的事务性处理 rabbitmq默认的处理方式为auto ack,这意味着当你从消息队列取出一个消息时,ack自动发送,mq就会将消息删除。而为了保证消息的正确处理,我们需要将消息处理修改为手动确认的方式。 (1) sender的手工确认模式 首先将ConnectionFactory的模式设置为publisherConfirms,如下
connectionFactory.setPublisherConfirms(true);
之后设置rabbitTemplate的confirmCallback,如下:
rabbitTemplate.setConfirmCallback(new RabbitTemplate.ConfirmCallback() {
@Override
public void confirm(CorrelationData correlationData, boolean ack, String cause) {
if (!ack) {
log.info("send message failed: " + cause); //+ correlationData.toString());
throw new RuntimeException("send error " + cause);
}
}
});
(2) consume的手工确认模式 首先在queue创建中指定模式
channel.exchangeDeclare(exchange, "direct", true, false, null);
/**
* Declare a queue
* @see com.rabbitmq.client.AMQP.Queue.Declare
* @see com.rabbitmq.client.AMQP.Queue.DeclareOk
* @param queue the name of the queue
* @param durable true if we are declaring a durable queue (the queue will survive a server restart)
* @param exclusive true if we are declaring an exclusive queue (restricted to this connection)
* @param autoDelete true if we are declaring an autodelete queue (server will delete it when no longer in use)
* @param arguments other properties (construction arguments) for the queue
* @return a declaration-confirm method to indicate the queue was successfully declared
* @throws java.io.IOException if an error is encountered
*/
channel.queueDeclare(queue, true, false, false, null);
只有在消息处理成功后发送ack确认,或失败后发送nack使信息重新投递
if (detailRes.isSuccess()) {
channel.basicAck(delivery.getEnvelope().getDeliveryTag(), false);
} else {
log.info("send message failed: " + detailRes.getErrMsg());
channel.basicNack(delivery.getEnvelope().getDeliveryTag(), false, true);
}
4 自动重连机制 为了保证rabbitmq的高可用性,我们使用rabbitmq Cluster模式,并配合haproxy。这样,在一台机器down掉时或者网络发生抖动时,就会发生当前连接失败的情况,如果不对这种情况做处理,就会造成当前的服务不可用。 在spring-rabbitmq中,已实现了connection的自动重连,但是connection重连后,channel的状态并不正确。因此我们需要自己捕捉ShutdownSignalException异常,并重新生成channel。如下:
catch (ShutdownSignalException e) {
e.printStackTrace();
channel.close();
//recreate channel
consumer = buildQueueConsumer(connection, queue);
}
5 consumer线程池 在对消息处理的过程中,我们期望多线程并行执行来增加效率,因此对consumer做了一个线程池的封装。 线程池通过builder模式构造,需要准备如下参数:
//线程数量
int threadCount;
//处理间隔(每个线程处理完成后休息的时间)
long intervalMils;
//exchange及queue信息
String exchange;
String routingKey;
String queue;
//用户自定义处理接口
MessageProcess<T> messageProcess;
核心循环也较为简单,代码如下:
public void run() {
while (!stop) {
try {
//2
DetailRes detailRes = messageConsumer.consume();
if (infoHolder.intervalMils > 0) {
try {
Thread.sleep(infoHolder.intervalMils);
} catch (InterruptedException e) {
e.printStackTrace();
log.info("interrupt " + e);
}
}
if (!detailRes.isSuccess()) {
log.info("run error " + detailRes.getErrMsg());
}
} catch (Exception e) {
e.printStackTrace();
log.info("run exception " + e);
}
}
}
6 使用示例 最后,我们还是用一个例子做结。 (1) 定义model
//参考lombok
@Data
@AllArgsConstructor
@NoArgsConstructor
public class UserMessage {
int id;
String name;
}
(2) rabbitmq配置 配置我们使用@Configuration实现,如下:
@Configuration
public class RabbitMQConf {
@Bean
ConnectionFactory connectionFactory() {
CachingConnectionFactory connectionFactory = new CachingConnectionFactory("127.0.0.1", 5672);
connectionFactory.setUsername("guest");
connectionFactory.setPassword("guest");
connectionFactory.setPublisherConfirms(true); // enable confirm mode
return connectionFactory;
}
}
(3) sender示例
@Service
public class SenderExample {
private static final String EXCHANGE = "example";
private static final String ROUTING = "user-example";
private static final String QUEUE = "user-example";
@Autowired
ConnectionFactory connectionFactory;
private MessageSender messageSender;
@PostConstruct
public void init() throws IOException, TimeoutException {
MQAccessBuilder mqAccessBuilder = new MQAccessBuilder(connectionFactory);
messageSender = mqAccessBuilder.buildMessageSender(EXCHANGE, ROUTING, QUEUE);
}
public DetailRes send(UserMessage userMessage) {
return messageSender.send(userMessage);
}
}
(4) MessageProcess(用户自定义处理接口)示例,本例中我们只是简单的将信息打印出来
public class UserMessageProcess implements MessageProcess<UserMessage> {
@Override
public DetailRes process(UserMessage userMessage) {
System.out.println(userMessage);
return new DetailRes(true, "");
}
}
(5) consumer示例
@Service
public class ConsumerExample {
private static final String EXCHANGE = "example";
private static final String ROUTING = "user-example";
private static final String QUEUE = "user-example";
@Autowired
ConnectionFactory connectionFactory;
private MessageConsumer messageConsumer;
@PostConstruct
public void init() throws IOException, TimeoutException {
MQAccessBuilder mqAccessBuilder = new MQAccessBuilder(connectionFactory);
messageConsumer = mqAccessBuilder.buildMessageConsumer(EXCHANGE, ROUTING, QUEUE, new UserMessageProcess());
}
public DetailRes consume() {
return messageConsumer.consume();
}
}
(6) 线程池consumer示例 在main函数中,我们使用一个独立线程发送数据,并使用线程池接收数据。
@Service
public class PoolExample {
private static final String EXCHANGE = "example";
private static final String ROUTING = "user-example";
private static final String QUEUE = "user-example";
@Autowired
ConnectionFactory connectionFactory;
private ThreadPoolConsumer<UserMessage> threadPoolConsumer;
@PostConstruct
public void init() {
MQAccessBuilder mqAccessBuilder = new MQAccessBuilder(connectionFactory);
MessageProcess<UserMessage> messageProcess = new UserMessageProcess();
threadPoolConsumer = new ThreadPoolConsumer.ThreadPoolConsumerBuilder<UserMessage>()
.setThreadCount(Constants.THREAD_COUNT).setIntervalMils(Constants.INTERVAL_MILS)
.setExchange(EXCHANGE).setRoutingKey(ROUTING).setQueue(QUEUE)
.setMQAccessBuilder(mqAccessBuilder).setMessageProcess(messageProcess)
.build();
}
public void start() throws IOException {
threadPoolConsumer.start();
}
public void stop() {
threadPoolConsumer.stop();
}
public static void main(String[] args) throws IOException {
ApplicationContext ac = new ClassPathXmlApplicationContext("applicationContext.xml");
PoolExample poolExample = ac.getBean(PoolExample.class);
final SenderExample senderExample = ac.getBean(SenderExample.class);
poolExample.start();
new Thread(new Runnable() {
int id = 0;
@Override
public void run() {
while (true) {
senderExample.send(new UserMessage(id++, "" + System.nanoTime()));
try {
Thread.sleep(1000L);
} catch (InterruptedException e) {
e.printStackTrace();
}
}
}
}).start();
}
}
7 github地址,路过的帮忙点个星星,谢谢^_^。
附: rabbitmq安装过程: mac版安装可以使用homebrew。brew install就可以,安装好之后通过brew services start rabbitmq启动服务。通过
have fun