RabbitMQ：The channelMax limit is reached. Try later.

​MDC

这个问题是我当初写项目时遇到的，因为用RabbitMQ做削峰处理，高并发情况下，channel数到达了限制，所以不能继续创建，相信大家也遇到过。

​ 正常来说，这个错误还是比较少见的，只不过项目需要保证消息的可靠性，所以采取了发送确认和消费手动确认机制，导致并发性能下降，从而出现这个问题。、

​ 这里先上结论，方便着急的小伙伴们改bug。

​ 结论：RabbitMQ java客户端在创建连接时，会向服务端发送一个请求，这个请求会获取到服务端的channelMax值，java客户端会自己进行一个处理，两者都不为0时，会选择一个小的值，如果你没有在rabbitmq.conf文件中修改channel_Max的值，那么java客户端会采用默认的2047或更小，这就会导致你明明在客户端连接上配置了channelMax（比如你配置了4095），但依旧会报错，而且web管理页面最大值依旧是2047

第一次修改配置不生效

出现这种情况经常伴随着消息丢失，而且消息丢失情况非常严重，达到了百分之二十的丢失率，这个丢失率也会因为并发量、每次消费数量等等配置的不同而变化。

由于项目是基于SpringBoot2.2的，yml暂时无法配置RequestChannelMax的值，这里只能采用直接通过set的方式放入值。

@Configuration
@Slf4j
public class RabbitMQConfig {


    @Bean
    public RabbitTemplate rabbitTemplate(final ConnectionFactory connectionFactory) {
        CachingConnectionFactory cachingConnectionFactory = (CachingConnectionFactory) connectionFactory;
        //这里我明明设置了4095，但是项目运行之后，压测之后，还是会报异常，而且报异常的时候，RabbitMQ           //web管理页面上的channel数依旧是2047，不得已只能分析源码了
        cachingConnectionFactory.getRabbitConnectionFactory().setRequestedChannelMax(4095);
        final RabbitTemplate rabbitTemplate = new RabbitTemplate(cachingConnectionFactory);
        rabbitTemplate.setMessageConverter(jackson2MessageConverter());
        rabbitTemplate.setMandatory(true);
        rabbitTemplate.setConfirmCallback((correlationData, b, s) -> {
            if(!b){
                log.error("confirmCallBack 发送失败的数据：{}",correlationData);
                log.error("confirmCallBack 确认情况：{}",b);
                log.error("confirmCallBack 发送失败的原因：{}",s);
            }
        });
        rabbitTemplate.setReturnCallback((message, i, s, s1, s2) -> {
            log.error("returnCallBack 消息：{}",message);
            log.error("returnCallBack 回应码：{}",i);
            log.error("returnCallBack 回应信息：{}",s);
            log.error("returnCallBack 交换机：{}",s1);
            log.error("returnCallBack 路由键：{}",s2);
        });
        return rabbitTemplate;
    }

    @Bean
    public Jackson2JsonMessageConverter jackson2MessageConverter() {
        return new Jackson2JsonMessageConverter();
    }
}

分析源码

首先是模拟出报错的场景，然后进入报异常的类。

发现是this.delegate.createChannel();方法返回的是一个空channel对象，进入这个方法看一下。

发现有一个ChannelManager对象，顾名思义，就是一个channel管理器，由它负责创建channel，那么看一下这个对象都有什么值呢？

com.rabbitmq.client.impl.ChannelManager#createChannel(com.rabbitmq.client.impl.AMQConnection)

public Channel createChannel() throws IOException {
        this.ensureIsOpen();
        ChannelManager cm = this._channelManager;
        if (cm == null) {
            return null;
        } else {
            Channel channel = cm.createChannel(this);
            this.metricsCollector.newChannel(channel);
            return channel;
        }
    }

只截取了部分代码，首先可以看到有一个int类型的channelMax，这个值就是channel的最大值，还有一个构造器，很明显，这个值是通过构造器传进来的，通过容器初始化时打断点进行跟踪，发现此时的channelMax依旧是2047，这也进一步证明了，值的覆盖或者处理发生在这个类调用之前。

com.rabbitmq.client.impl.ChannelManager

public class ChannelManager {
    private static final Logger LOGGER = LoggerFactory.getLogger(ChannelManager.class);
    private final Object monitor;
    private final Map<Integer, ChannelN> _channelMap;
    private final IntAllocator channelNumberAllocator;
    private final ConsumerWorkService workService;
    private final Set<CountDownLatch> shutdownSet;
    private final int _channelMax;
    private ExecutorService shutdownExecutor;
    private final ThreadFactory threadFactory;
    private int channelShutdownTimeout;
    protected final MetricsCollector metricsCollector;
    
    public ChannelManager(ConsumerWorkService workService, int channelMax, ThreadFactory threadFactory, MetricsCollector metricsCollector) {
        this.monitor = new Object();
        this._channelMap = new HashMap();
        this.shutdownSet = new HashSet();
        this.channelShutdownTimeout = 63000;
        if (channelMax == 0) {
            channelMax = 65535;
        }

        this._channelMax = channelMax;
        this.channelNumberAllocator = new IntAllocator(1, channelMax);
        this.workService = workService;
        this.threadFactory = threadFactory;
        this.metricsCollector = metricsCollector;
    }
}

进一步跟踪之后，发现在AMQConnection类里的instantiateChannelManager()方法调用了构造器，继续往上追踪。

com.rabbitmq.client.impl.AMQConnection#instantiateChannelManager
protected ChannelManager instantiateChannelManager(int channelMax, ThreadFactory threadFactory) {
        ChannelManager result = new ChannelManager(this._workService, channelMax, threadFactory, this.metricsCollector);
        this.configureChannelManager(result);
        return result;
    }

在AMQConnetion类的start()方法中最终发现了值改变的地方。

this.requestedChannelMax值是我在配置类中配置的4095

connTune.getChannelMax()是2047

也就是说，negotiateChannelMax()方法对这两个值进行了处理，最终选择了2047

            int channelMax = this.negotiateChannelMax(this.requestedChannelMax, connTune.getChannelMax());
            this._channelManager = this.instantiateChannelManager(channelMax, this.threadFactory);

最终发现这么一段处理逻辑，如果两个数字都不为0，那么就取最小的，反之取最大的，看到这里是明白做了什么处理，但是还是有一处不明白，2047的值究竟从何处来的？

com.rabbitmq.client.impl.AMQConnection#negotiateChannelMax
    /**
     * Private API, allows for easier simulation of bogus clients.
     */
    protected int negotiateChannelMax(int requestedChannelMax, int serverMax) {
        return negotiatedMaxValue(requestedChannelMax, serverMax);
    }


    private static int negotiatedMaxValue(int clientValue, int serverValue) {
        return (clientValue == 0 || serverValue == 0) ?
            Math.max(clientValue, serverValue) :
            Math.min(clientValue, serverValue);
    }

通过对connTune的追寻，发现了这段处理，debug也证明了确实在这里获取的2047这个值，

其实不管从方法名rpc()还是变量名serverResponse来看，这个都是做了一个请求，那么向谁请求其实很显而易见了，这里向RabbitMQ端做了一个请求，用来索取MQ端的channelMax、frameMax、heartBeat值等等

Tune connTune = null;

        try {
           ......
                try {
                    Method serverResponse = this._channel0.rpc((Method)method, this.handshakeTimeout / 2).getMethod();
                    if (serverResponse instanceof Tune) {
                        connTune = (Tune)serverResponse;
                    }
                    ......

到现在其实就很明确了，我们只在客户端修改边界值配置是无效的，必须同步修改MQ服务端的配置，也就是rabbitmq.conf文件

## Set the max permissible number of channels per connection.
## 0 means "no limit".
##在配置文件中，输入以下参数和自己想要设置的值即可，如果用不到2047，那就不用配置
# channel_max = 128

其实问题并不大，主要还是不了解MQ的一个客户端连接过程，导致耗费了大量时间。

这里还是推荐大家，先用百度搜索，第一页看不到正确解决方案，那就去StackOverflow网站，还不行的话，那就使用终极大法，要么官网逐行看文档，要么走一波源码，也是锻炼自己解决问题的思路和能力。

https://blog.csdn.net/qq_35374224/article/details/106721801


https://github.com/spring-projects/spring-amqp/issues/853

2020-09-09 14:02:36.549  INFO 6 --- [org.springframework.amqp.rabbit.RabbitListenerEndpointContainer#10-60623] c.r.c.impl.AbstractMetricsCollector      : Error while computing metrics in newChannel: null
2020-09-09 14:02:36.549 ERROR 6 --- [org.springframework.amqp.rabbit.RabbitListenerEndpointContainer#10-60623] o.s.a.r.l.SimpleMessageListenerContainer : Failed to check/redeclare auto-delete queue(s).

org.springframework.amqp.AmqpResourceNotAvailableException: The channelMax limit is reached. Try later.
        at org.springframework.amqp.rabbit.connection.SimpleConnection.createChannel(SimpleConnection.java:60) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory$ChannelCachingConnectionProxy.createBareChannel(CachingConnectionFactory.java:1434) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory$ChannelCachingConnectionProxy.access$200(CachingConnectionFactory.java:1420) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory.doCreateBareChannel(CachingConnectionFactory.java:723) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory.createBareChannel(CachingConnectionFactory.java:706) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory.getCachedChannelProxy(CachingConnectionFactory.java:676) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory.getChannel(CachingConnectionFactory.java:567) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory.access$1600(CachingConnectionFactory.java:102) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.connection.CachingConnectionFactory$ChannelCachingConnectionProxy.createChannel(CachingConnectionFactory.java:1439) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.core.RabbitTemplate.doExecute(RabbitTemplate.java:2095) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.core.RabbitTemplate.execute(RabbitTemplate.java:2062) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.core.RabbitTemplate.execute(RabbitTemplate.java:2042) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.core.RabbitAdmin.getQueueInfo(RabbitAdmin.java:407) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.core.RabbitAdmin.getQueueProperties(RabbitAdmin.java:391) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.listener.AbstractMessageListenerContainer.attemptDeclarations(AbstractMessageListenerContainer.java:1842) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.listener.AbstractMessageListenerContainer.redeclareElementsIfNecessary(AbstractMessageListenerContainer.java:1823) ~[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.listener.SimpleMessageListenerContainer$AsyncMessageProcessingConsumer.initialize(SimpleMessageListenerContainer.java:1349)[spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at org.springframework.amqp.rabbit.listener.SimpleMessageListenerContainer$AsyncMessageProcessingConsumer.run(SimpleMessageListenerContainer.java:1195) [spring-rabbit-2.2.10.RELEASE.jar!/:2.2.10.RELEASE]
        at java.lang.Thread.run(Thread.java:748) [na:1.8.0_144]

2020-09-09 14:02:36.549  INFO 6 --- [org.springframework.amqp.rabbit.RabbitListenerEndpointContainer#10-60623] c.r.c.impl.AbstractMetricsCollector      : Error while computing metrics in newChannel: null