springboot RabbitMQ客户端连接故障恢复

最近做RabbitMQ故障演练发现RabbitMQ服务器停止后，基于springboot的消费端不可以自动的恢复，队列的消费者消失，消息一直积压到队列中，这种情况肯定是不可接收的；通过研究源代码找到了解决方案。

一、添加自动恢复配置automaticRecovery

 CachingConnectionFactory factory = new CachingConnectionFactory(connectionFactory);
cachingConnectionFactoryConfigurer.configure(factory);//设置TCP连接超时时间，默认：60000ms
factory.getRabbitConnectionFactory().setConnectionTimeout(properties.getConnectionTimeout());
//启用或禁用连接自动恢复，默认：false
factory.getRabbitConnectionFactory().setAutomaticRecoveryEnabled(properties.isAutomaticRecovery());
//设置连接恢复时间间隔，默认：5000ms
factory.getRabbitConnectionFactory().setNetworkRecoveryInterval(properties.getNetworkRecoveryInterval());
//启用或禁用拓扑恢复，默认：true【拓扑恢复功能可以帮助消费者重新声明之前定义的队列、交换机和绑定等拓扑结构】
factory.getRabbitConnectionFactory().setTopologyRecoveryEnabled(properties.isTopologyRecovery());
//替换默认异常处理DefaultExceptionHandler
factory.getRabbitConnectionFactory().setExceptionHandler(new DefaultMqExceptionHandler());
//添加连接监听器
factory.addConnectionListener(new DefaultMqConnectionListener(factory));

通过上述配置如果RabbitMQ服务器发生故障，则会自动重启恢复连接及队列的消费者，如果恢复失败则会间隔5000ms再次重试；在这里提一个问题，如果服务重试一直失败，重试的上限是多少？带着这个问题我们分析下源码。

二、RabbitMQ客户端实现连接的自动恢复功能

AutorecoveringConnection#beginAutomaticRecovery是在 RabbitMQ 客户端库层面实现的连接的自动恢复功能。当 RabbitMQ 连接出现故障时，它会尝试重新建立连接，以确保消息传递的可靠性。

  private synchronized void beginAutomaticRecovery() throws InterruptedException {//获取故障恢复连接的间隔时间，实际是设置的：networkRecoveryIntervalfinal long delay = this.params.getRecoveryDelayHandler().getDelay(0);if (delay > 0)  {//等待指定的间隔时间this.wait(delay);}//调用恢复通知监听器this.notifyRecoveryListenersStarted();//获取恢复建立的连接对象final RecoveryAwareAMQConnection newConn = this.recoverConnection();//如果为null则直接返回if (newConn == null) {return;}//连接已经恢复建立，恢复监听器、channel等资源LOGGER.debug("Connection {} has recovered", newConn);this.addAutomaticRecoveryListener(newConn);this.recoverShutdownListeners(newConn);this.recoverBlockedListeners(newConn);this.recoverChannels(newConn);// don't assign new delegate connection until channel recovery is completethis.delegate = newConn;//判断是否恢复拓扑结构，如果开启则开启拓扑结构恢复if (this.params.isTopologyRecoveryEnabled()) {notifyTopologyRecoveryListenersStarted();recoverTopology(params.getTopologyRecoveryExecutor());}this.notifyRecoveryListenersComplete();}

addAutomaticRecoveryListener自动恢复监听器

private void addAutomaticRecoveryListener(final RecoveryAwareAMQConnection newConn) {final AutorecoveringConnection c = this;// this listener will run after shutdown listeners,// see https://github.com/rabbitmq/rabbitmq-java-client/issues/135RecoveryCanBeginListener starter = cause -> {try {if (shouldTriggerConnectionRecovery(cause)) {//开始自动回复c.beginAutomaticRecovery();}} catch (Exception e) {newConn.getExceptionHandler().handleConnectionRecoveryException(c, e);}};synchronized (this) {newConn.addRecoveryCanBeginListener(starter);}
}

init初始化

public void init() throws IOException, TimeoutException {//建立连接，否则抛出异常this.delegate = this.cf.newConnection();//自动回复监听器this.addAutomaticRecoveryListener(delegate);
}

三、消费端实现消息的消费和处理

SimpleMessageListenerContainer.AsyncMessageProcessingConsumer#run是应用程序层面实现消息的消费和处理，它负责从RabbitMQ中接收消息并进行相应的逻辑处理：

@Override // NOSONAR - complexity - many catch blocks
public void run() { // NOSONAR - line count...try {//消费端初始化方法initialize();//当消费端是活跃状态，或者队列非空，或者消费端未被关闭则进入主循环while (isActive(this.consumer) || this.consumer.hasDelivery() || !this.consumer.cancelled()) {mainLoop();}}catch (InterruptedException e) {...}
}

消费端initialize初始化方法：

	private void initialize() throws Throwable { // NOSONARtry {redeclareElementsIfNecessary();//启动消费端初始化this.consumer.start();this.start.countDown();}catch (QueuesNotAvailableException e) {if (isMissingQueuesFatal()) {throw e;}else {this.start.countDown();//消费端启动异常等待处理handleStartupFailure(this.consumer.getBackOffExecution());throw e;}}catch (FatalListenerStartupException ex) {if (isPossibleAuthenticationFailureFatal()) {throw ex;}else {Throwable possibleAuthException = ex.getCause().getCause();if (!(possibleAuthException instanceof PossibleAuthenticationFailureException)) {throw ex;}else {this.start.countDown();//消费端启动异常等待处理handleStartupFailure(this.consumer.getBackOffExecution());throw possibleAuthException;}}}catch (Throwable t) { //NOSONARthis.start.countDown();//消费端启动异常等待处理handleStartupFailure(this.consumer.getBackOffExecution());throw t;}if (getTransactionManager() != null) {/** Register the consumer's channel so it will be used by the transaction manager* if it's an instance of RabbitTransactionManager.*/ConsumerChannelRegistry.registerConsumerChannel(this.consumer.getChannel(), getConnectionFactory());}}

消费端异常等待处理处理：

protected void handleStartupFailure(BackOffExecution backOffExecution) {//获取等待时间间隔，参考FixedBackOff类实现long recoveryInterval = backOffExecution.nextBackOff();if (BackOffExecution.STOP == recoveryInterval) {synchronized (this) {if (isActive()) {logger.warn("stopping container - restart recovery attempts exhausted");stop();}}return;}try {if (logger.isDebugEnabled() && isActive()) {logger.debug("Recovering consumer in " + recoveryInterval + " ms.");}//当前时间加上等待时间long timeout = System.currentTimeMillis() + recoveryInterval;//如果当前时间小于等待时间，则休眠200毫秒，再次尝试while (isActive() && System.currentTimeMillis() < timeout) {Thread.sleep(RECOVERY_LOOP_WAIT_TIME);}}catch (InterruptedException e) {Thread.currentThread().interrupt();throw new IllegalStateException("Irrecoverable interruption on consumer restart", e);}}

FixedBackOff回退等待时间类实现：

public class FixedBackOff implements BackOff {// 默认恢复重试间隔public static final long DEFAULT_INTERVAL = 5000L;//最大重试次数，可以认为无限大public static final long UNLIMITED_ATTEMPTS = Long.MAX_VALUE;// 默认恢复重试间隔private long interval = 5000L;//最大重试次数，可以认为无限大private long maxAttempts = Long.MAX_VALUE;public FixedBackOff() {}public FixedBackOff(long interval, long maxAttempts) {this.interval = interval;this.maxAttempts = maxAttempts;}public void setInterval(long interval) {this.interval = interval;}public long getInterval() {return this.interval;}public void setMaxAttempts(long maxAttempts) {this.maxAttempts = maxAttempts;}public long getMaxAttempts() {return this.maxAttempts;}public BackOffExecution start() {return new FixedBackOffExecution();}private class FixedBackOffExecution implements BackOffExecution {private long currentAttempts;private FixedBackOffExecution() {this.currentAttempts = 0L;}//获取下一次尝试的时间间隔，可以认为一直都是5000mspublic long nextBackOff() {++this.currentAttempts;return this.currentAttempts <= FixedBackOff.this.getMaxAttempts() ? FixedBackOff.this.getInterval() : -1L;}public String toString() {String attemptValue = FixedBackOff.this.maxAttempts == Long.MAX_VALUE ? "unlimited" : String.valueOf(FixedBackOff.this.maxAttempts);return "FixedBackOff{interval=" + FixedBackOff.this.interval + ", currentAttempts=" + this.currentAttempts + ", maxAttempts=" + attemptValue + '}';}}
}

总结：综上源码分析可知消费端故障恢复重试等待时间是5000ms,重试次数可以认为是无限制（Long最大值）

mainloop主循环逻辑：

		private void mainLoop() throws Exception { // NOSONAR Exceptiontry {if (SimpleMessageListenerContainer.this.stopNow.get()) {this.consumer.forceCloseAndClearQueue();return;}//接收客户端发送过来的消息，至少获取一条boolean receivedOk = receiveAndExecute(this.consumer); // At least one message receivedif (SimpleMessageListenerContainer.this.maxConcurrentConsumers != null) {checkAdjust(receivedOk);}long idleEventInterval = getIdleEventInterval();if (idleEventInterval > 0) {if (receivedOk) {updateLastReceive();}else {long now = System.currentTimeMillis();long lastAlertAt = SimpleMessageListenerContainer.this.lastNoMessageAlert.get();long lastReceive = getLastReceive();if (now > lastReceive + idleEventInterval&& now > lastAlertAt + idleEventInterval&& SimpleMessageListenerContainer.this.lastNoMessageAlert.compareAndSet(lastAlertAt, now)) {publishIdleContainerEvent(now - lastReceive);}}}}catch (ListenerExecutionFailedException ex) {// Continue to process, otherwise re-throwif (ex.getCause() instanceof NoSuchMethodException) {throw new FatalListenerExecutionException("Invalid listener", ex);}}catch (AmqpRejectAndDontRequeueException rejectEx) {/**  These will normally be wrapped by an LEFE if thrown by the*  listener, but we will also honor it if thrown by an*  error handler.*/}}

receiveAndExecute接收和处理消息：

private boolean receiveAndExecute(final BlockingQueueConsumer consumer) throws Exception { // NOSONARPlatformTransactionManager transactionManager = getTransactionManager();if (transactionManager != null) {try {if (this.transactionTemplate == null) {this.transactionTemplate =new TransactionTemplate(transactionManager, getTransactionAttribute());}return this.transactionTemplate.execute(status -> { // NOSONAR null never returnedRabbitResourceHolder resourceHolder = ConnectionFactoryUtils.bindResourceToTransaction(new RabbitResourceHolder(consumer.getChannel(), false),getConnectionFactory(), true);// unbound in ResourceHolderSynchronization.beforeCompletion()try {//接收处理消息return doReceiveAndExecute(consumer);}catch (RuntimeException e1) {prepareHolderForRollback(resourceHolder, e1);throw e1;}catch (Exception e2) {throw new WrappedTransactionException(e2);}});}catch (WrappedTransactionException e) { // NOSONAR exception flow controlthrow (Exception) e.getCause();}}//接收处理消息return doReceiveAndExecute(consumer);}

调用具体的消息监听器消费消息：

	private void doExecuteListener(Channel channel, Object data) {if (data instanceof Message) {Message message = (Message) data;if (this.afterReceivePostProcessors != null) {for (MessagePostProcessor processor : this.afterReceivePostProcessors) {message = processor.postProcessMessage(message);if (message == null) {throw new ImmediateAcknowledgeAmqpException("Message Post Processor returned 'null', discarding message");}}}if (this.deBatchingEnabled && this.batchingStrategy.canDebatch(message.getMessageProperties())) {this.batchingStrategy.deBatch(message, fragment -> invokeListener(channel, fragment));}else {invokeListener(channel, message);}}else {invokeListener(channel, data);}}

GitHub代码：https://github.com/mingyang66/spring-parent