前言
目前有两套RocketMQ集群,集群A包含topic
名称为cluster_A_topic
,集群B包含topic
名称为cluster_B_topic
,在应用服务OrderApp
上通过RocketMQ Client
创建两个DefaultMQProducer
实例发送消息给集群A和集群B,架构图如下:
根据上述架构图,我们给出的示例代码如下:
// 创建第一个DefaultMQProducer DefaultMQProducer producer1 = new DefaultMQProducer("producer_group_1"); // 设置nameServer地址 producer1.setNamesrvAddr("192.168.2.230:9876"); try { producer1.start(); // 发送消息 SendResult result1 = producer1.send(new Message("cluster_A_topic", "ping".getBytes(StandardCharsets.UTF_8))); switch (result1.getSendStatus()) { case SEND_OK: System.out.println("cluster_A_topic 发送成功!"); break; case FLUSH_DISK_TIMEOUT: System.out.println("cluster_A_topic 持久化失败!"); break; case FLUSH_SLAVE_TIMEOUT: System.out.println("cluster_A_topic 同步slave失败!"); break; case SLAVE_NOT_AVAILABLE: System.out.println("cluster_A_topic 副本不可用!"); } } catch (Exception e) { e.printStackTrace(); } // 创建第二个DefaultMQProducer DefaultMQProducer producer2 = new DefaultMQProducer("producer_group_2"); // 设置nameServer地址 producer2.setNamesrvAddr("192.168.2.231:9876"); try { producer2.start(); // 发送消息 SendResult result2 = producer2.send(new Message("cluster_B_topic", "ping".getBytes(StandardCharsets.UTF_8))); switch (result2.getSendStatus()) { case SEND_OK: System.out.println("cluster_B_topic 发送成功!"); break; case FLUSH_DISK_TIMEOUT: System.out.println("cluster_B_topic 持久化失败!"); break; case FLUSH_SLAVE_TIMEOUT: System.out.println("cluster_B_topic 同步slave失败!"); break; case SLAVE_NOT_AVAILABLE: System.out.println("cluster_B_topic 副本不可用!"); } return "ok"; } catch (Exception e) { e.printStackTrace(); } finally { producer1.shutdown(); producer2.shutdown(); } 复制代码
结果竟然报错了,报错内容时cluster_B_topic
不存在:
经过不断的测试,发现只有放在最前面启动的DefaultMQProducer
会生效,后面启动的DefaultMQProducer
发送消息就报错说对应的topic
不存在,而且报错的broker
竟然是前面启动的DefaultMQProducer
对应的broker
。这就不科学了,难道RocketMQ不允许在一个应用上创建多个生产者?
问题定位
首先说明一下,当前使用的RocketMQ Client
版本是4.8.0
。为了确定是哪儿出了问题,不得不对源码来一波探索[哭泣脸😢]。
我们都知道生产者是发送消息给Broker
的,获取Broker
信息是通过连接NameServer
获取的。既然报错的Broker
和目标Broker
竟然不对应,肯定是后面启动的生产者获取的Broker
不对。有了最基本的判断,我们先从DefaultMQProducer#start()
入手,最终我们定位到这样一段代码DefaultMQProducerImpl#start(final boolean startFactory)
:
public void start(final boolean startFactory) throws MQClientException { switch (this.serviceState) { case CREATE_JUST: this.serviceState = ServiceState.START_FAILED; this.checkConfig(); // 如果生产者group名称不是`CLIENT_INNER_PRODUCER`,那么修改InstanceName值 if (!this.defaultMQProducer.getProducerGroup().equals(MixAll.CLIENT_INNER_PRODUCER_GROUP)) { this.defaultMQProducer.changeInstanceNameToPID(); } // 创建MQClientInstance实例 this.mQClientFactory = MQClientManager.getInstance().getOrCreateMQClientInstance(this.defaultMQProducer, rpcHook); // 注册生产者实例到MQClientInstance中 boolean registerOK = mQClientFactory.registerProducer(this.defaultMQProducer.getProducerGroup(), this); if (!registerOK) { this.serviceState = ServiceState.CREATE_JUST; throw new MQClientException("The producer group[" + this.defaultMQProducer.getProducerGroup() + "] has been created before, specify another name please." + FAQUrl.suggestTodo(FAQUrl.GROUP_NAME_DUPLICATE_URL), null); } // 添加TBW102对应的topic信息,broker设置autoCreateTopicEnable = true才起作用 this.topicPublishInfoTable.put(this.defaultMQProducer.getCreateTopicKey(), new TopicPublishInfo()); if (startFactory) { // 启动刚刚创建的MQClientInstance实例 mQClientFactory.start(); } log.info("the producer [{}] start OK. sendMessageWithVIPChannel={}", this.defaultMQProducer.getProducerGroup(), this.defaultMQProducer.isSendMessageWithVIPChannel()); // 修改服务状态为RUNNING this.serviceState = ServiceState.RUNNING; break; case RUNNING: case START_FAILED: case SHUTDOWN_ALREADY: throw new MQClientException("The producer service state not OK, maybe started once, " + this.serviceState + FAQUrl.suggestTodo(FAQUrl.CLIENT_SERVICE_NOT_OK), null); default: break; } 复制代码
上面的代码主要是创建了MQClientInstance
实例,并且通过start()
方法启动。
通过针对这两段代码的debug,我们发现创建的两个DefaultMQProducer
对象是共用了一个MQClientInstance
实例,并且所有针对NameServer
和Broker
的远程操作全部是通过MQClientInstance
实例来做的。比如发送消息的时候需要找到对应的Broker
下的消息队列:
private TopicPublishInfo tryToFindTopicPublishInfo(final String topic) { TopicPublishInfo topicPublishInfo = this.topicPublishInfoTable.get(topic); if (null == topicPublishInfo || !topicPublishInfo.ok()) { this.topicPublishInfoTable.putIfAbsent(topic, new TopicPublishInfo()); // 从NameServer更新topic路由 this.mQClientFactory.updateTopicRouteInfoFromNameServer(topic); topicPublishInfo = this.topicPublishInfoTable.get(topic); } if (topicPublishInfo.isHaveTopicRouterInfo() || topicPublishInfo.ok()) { return topicPublishInfo; } else { this.mQClientFactory.updateTopicRouteInfoFromNameServer(topic, true, this.defaultMQProducer); topicPublishInfo = this.topicPublishInfoTable.get(topic); return topicPublishInfo; } } 复制代码
最终我们发现两个DefaultMQProducer
对象都是去同一个NameServer
下获取对应的topic
信息,这下问题就定位到了:因为使用了同一个MQClientInstance
实例导致不同的DefaultMQProducer
去访问了同一个NameServer
,同一个集群需要同时接收两个topic
的消息,也就出现了前面的报错说topic
不存在的情况。
如何解决
我们来看看MQClientInstance
实例是如何保证唯一性的:
public MQClientInstance getOrCreateMQClientInstance(final ClientConfig clientConfig, RPCHook rpcHook) { // 生成clientID String clientId = clientConfig.buildMQClientId(); // 从缓存中获取MQClientInstance MQClientInstance instance = this.factoryTable.get(clientId); if (null == instance) { // 没有缓存的话就创建一个MQClientInstance instance = new MQClientInstance(clientConfig.cloneClientConfig(), this.factoryIndexGenerator.getAndIncrement(), clientId, rpcHook); // 新创建出来的再放进缓存 MQClientInstance prev = this.factoryTable.putIfAbsent(clientId, instance); if (prev != null) { instance = prev; log.warn("Returned Previous MQClientInstance for clientId:[{}]", clientId); } else { log.info("Created new MQClientInstance for clientId:[{}]", clientId); } } // 返回MQClientInstance实例 return instance; } 复制代码
我们之所以拿到的MQClientInstance
实例是同一个,是因为在同一个服务下创建的clientId
相同:
public String buildMQClientId() { StringBuilder sb = new StringBuilder(); sb.append(this.getClientIP()); sb.append("@"); sb.append(this.getInstanceName()); if (!UtilAll.isBlank(this.unitName)) { sb.append("@"); sb.append(this.unitName); } return sb.toString(); } 复制代码
两个clientId
都是192.168.18.173@14933
,为了防止clientId
相同,我们可以在创建DefaultMQProducer
实例是加上unitName
值,保证两个unitName
值不同来避免共享同一个MQClientInstance
。
DefaultMQProducer producer1 = new DefaultMQProducer("producer_group_1"); producer1.setNamesrvAddr("192.168.2.230:9876"); producer1.setUnitName("producer1"); producer1.start(); DefaultMQProducer producer2 = new DefaultMQProducer("producer_group_1"); producer2.setNamesrvAddr("192.168.2.231:9876"); producer2.setUnitName("producer2"); producer2.start(); 复制代码
通过上述代码修改后,两个消息都发送成功了。
另一个办法就是升级RocketMQ Client
到4.9.0
,我们来看一下RocketMQ Client 4.9.0
是怎么解决这个问题的:
public void changeInstanceNameToPID() { if (this.instanceName.equals("DEFAULT")) { this.instanceName = UtilAll.getPid() + "#" + System.nanoTime(); } } 复制代码
RocketMQ Client 4.9.0
在后面补充了一个纳秒值,之前的代码是这样的:
public void changeInstanceNameToPID() { if (this.instanceName.equals("DEFAULT")) { this.instanceName = String.valueOf(UtilAll.getPid()); } } 复制代码
也就是说,在新的版本中,一个应用服务内创建多个DefaultMQProducer
就会有多个MQClientInstance
实例对应,不会再出现我们前面的报错。