🏆【Alibaba中间件技术系列】「RocketMQ技术专题」分

RocketMQ消费失败重试机制分析

今天我们分析一下RocketMQ消费重试机制，如果执行以下的RocketMQ的消费服务的代码

try {

            try {

                if (messageExtWrappers.size() > 0) {

                    try {

                        var22 = messageExtWrappers.iterator();

                        while(var22.hasNext()) {

                            messageExt = (MessageExt)var22.next();

                        }

                    } catch (Throwable var16) {

                        ;

                    }

                    this.consume(messageExtWrappers, context);

                }

                LOGGER.info("MQ_CON_SUCCESS {} BROKER {} QUEUE {}", new Object[]{topic, broker, queueId});

                ConsumeConcurrentlyStatus var23 = ConsumeConcurrentlyStatus.CONSUME_SUCCESS;

                return var23;

            } catch (MessageListenerConcurrentlyException var17) {

                LOGGER.error("MQ_CON_EX {} BROKER {} QUEUE {}", new Object[]{topic, broker, queueId, var17});

                throw var17;

            } catch (Throwable var18) {

                LOGGER.error("MQ_CON_EX {} BROKER {} QUEUE {}", new Object[]{topic, broker, queueId, var18});

                LOGGER.info("MQ_CON_RECONSUME {} BROKER {} QUEUE {}", new Object[]{topic, broker, queueId});

                if (CollectionUtils.isNotEmpty(msgs) && ((MessageExt)msgs.get(0)).getDelayTimeLevel() >= 2 + this.retryTimes) {

                    context.setDelayLevelWhenNextConsume(-1);

                }

            }

当进行消费数据后得出的结果日志如下：

第一次消费

MQ_CON_MSG topic MSG MessageExt [queueId=1, storeSize=453, queueOffset=25, sysFlag=0, bornTimestamp=1566785215908, bornHost=/10.42.0.77:54608, storeTimestamp=1566785215908, storeHost=/10.42.0.244:10911, msgId=0A2A00F400002A9F000000000B77CE84, commitLogOffset=192401028, bodyCRC=53737244, reconsumeTimes=0, preparedTransactionOffset=0, toString()=Message{topic='gmcf-lsc-zhongbang-repu-calc-from-topic', flag=0, properties={MIN_OFFSET=0, _catChildMessageId1=gmcf-lsc-job-0a2a004d-435218-15071, HASH_CODE=690132963, MAX_OFFSET=26, _catRootMessageId=lts-0a2a006b-434483-0, CONSUME_START_TIME=1566785215911, SERIALIZE_CLASS=java.lang.String, _catParentMessageId=lts-0a2a006b-435218-524, _catChildMessageId=gmcf-lsc-job-0a2a004d-435218-15072, UNIQ_KEY=0A2A004D000938AF386882EAA5A40112, WAIT=true}, body=[71, 77, 66, 69, 67, 49, 57, 48, 56, 50, 52, 49, 48, 52, 48, 52, 54, 56, 57, 52, 48, 52, 48, 56, 48], transactionId='null’}]

第一次retry（reconsumeTimes=1,DELAY=3）

MQ_CON_MSG %RETRY%consumerGroup MSG MessageExt [queueId=0, storeSize=608, queueOffset=1187, sysFlag=0, bornTimestamp=1566785215923, bornHost=/10.42.0.77:54608, storeTimestamp=1566785226241, storeHost=/10.42.0.244:10911, msgId=0A2A00F400002A9F000000000B785900, commitLogOffset=192436480, bodyCRC=893293938, reconsumeTimes=1, preparedTransactionOffset=0, toString()=Message{topic='gmcf-lsc-zhongbang-repu-calc-from-topic', flag=0, properties={_catChildMessageId1=gmcf-lsc-job-0a2a004d-435218-15075, _catRootMessageId=lts-0a2a006b-434483-0, CONSUME_START_TIME=1566785226242, SERIALIZE_CLASS=java.lang.String, _catParentMessageId=lts-0a2a006b-435218-524, _catChildMessageId=gmcf-lsc-job-0a2a004d-435218-15076, MIN_OFFSET=0, REAL_TOPIC=%RETRY%gmcf-lsc-consumerGroup, ORIGIN_MESSAGE_ID=0A2A00F400002A9F000000000B77D049, RETRY_TOPIC=gmcf-lsc-zhongbang-repu-calc-from-topic, HASH_CODE=1506102461, MAX_OFFSET=1188, UNIQ_KEY=0A2A004D000938AF386882EAA5B30113, WAIT=false, DELAY=3, REAL_QID=0}, body=[71, 77, 66, 69, 67, 49, 57, 48, 56, 50, 52, 49, 48, 52, 49, 49, 49, 50, 48, 55, 48, 52, 48, 56, 49], transactionId='null'}]

第二次retry（reconsumeTimes=2, DELAY=4）

MQ_CON_MSG %RETRY%consumerGroup MSG MessageExt [queueId=0, storeSize=608, queueOffset=1209, sysFlag=0, bornTimestamp=1566785215923, bornHost=/10.42.0.77:54608, storeTimestamp=1566785256680, storeHost=/10.42.0.244:10911, msgId=0A2A00F400002A9F000000000B791399, commitLogOffset=192484249, bodyCRC=893293938, reconsumeTimes=2, preparedTransactionOffset=0, toString()=Message{topic='gmcf-lsc-zhongbang-repu-calc-from-topic', flag=0, properties={_catChildMessageId1=gmcf-lsc-job-0a2a004d-435218-15075, _catRootMessageId=lts-0a2a006b-434483-0, CONSUME_START_TIME=1566785256728, SERIALIZE_CLASS=java.lang.String, _catParentMessageId=lts-0a2a006b-435218-524, _catChildMessageId=gmcf-lsc-job-0a2a004d-435218-15076, MIN_OFFSET=0, REAL_TOPIC=%RETRY%gmcf-lsc-consumerGroup, ORIGIN_MESSAGE_ID=0A2A00F400002A9F000000000B77D049, RETRY_TOPIC=gmcf-lsc-zhongbang-repu-calc-from-topic, HASH_CODE=1506102461, MAX_OFFSET=1210, UNIQ_KEY=0A2A004D000938AF386882EAA5B30113, WAIT=false, DELAY=4, REAL_QID=0}, body=[71, 77, 66, 69, 67, 49, 57, 48, 56, 50, 52, 49, 48, 52, 49, 49, 49, 50, 48, 55, 48, 52, 48, 56, 49], transactionId='null'}]

第三次retry（reconsumeTimes=3, DELAY=5）

MQ_CON_MSG %RETRY%consumerGroup MSG MessageExt [queueId=0, storeSize=608, queueOffset=1228, sysFlag=0, bornTimestamp=1566785215923, b

RocketMQ的默认的配置是messageDelayLevel=1s 5s 10s 30s 1m 2m 3m 4m 5m 6m 7m 8m 9m 10m 20m 30m 1h 2h，分别代表延迟level1-level18，为什么不是从1开始呢？

if (CollectionUtils.isNotEmpty(msgs) && ((MessageExt)msgs.get(0)).getDelayTimeLevel() >= 2 + this.retryTimes) {

        context.setDelayLevelWhenNextConsume(-1);

}

当重试达到满足条件的时候，不再重试，直接放到dlq队列里面。如果不控制的，会一直重试到最高DelayLevel 18。

从DefaultMQPullConsumerImpl类里面找到一段代码

public void sendMessageBack(MessageExt msg, int delayLevel, String brokerName, String consumerGroup) throws RemotingException, MQBrokerException, InterruptedException, MQClientException {

        try {

            String brokerAddr = null != brokerName ? this.mQClientFactory.findBrokerAddressInPublish(brokerName) : RemotingHelper.parseSocketAddressAddr(msg.getStoreHost());

            if (UtilAll.isBlank(consumerGroup)) {

                consumerGroup = this.defaultMQPullConsumer.getConsumerGroup();

            }

            this.mQClientFactory.getMQClientAPIImpl().consumerSendMessageBack(brokerAddr, msg, consumerGroup, delayLevel, 3000L, this.defaultMQPullConsumer.getMaxReconsumeTimes());

        } catch (Exception var8) {

            this.log.error("sendMessageBack Exception, " + this.defaultMQPullConsumer.getConsumerGroup(), var8);

            Message newMsg = new Message(MixAll.getRetryTopic(this.defaultMQPullConsumer.getConsumerGroup()), msg.getBody());

            String originMsgId = MessageAccessor.getOriginMessageId(msg);

            MessageAccessor.setOriginMessageId(newMsg, UtilAll.isBlank(originMsgId) ? msg.getMsgId() : originMsgId);

            newMsg.setFlag(msg.getFlag());

            MessageAccessor.setProperties(newMsg, msg.getProperties());

            MessageAccessor.putProperty(newMsg, "RETRY_TOPIC", msg.getTopic());

            MessageAccessor.setReconsumeTime(newMsg, String.valueOf(msg.getReconsumeTimes() + 1));

            MessageAccessor.setMaxReconsumeTimes(newMsg, String.valueOf(this.defaultMQPullConsumer.getMaxReconsumeTimes()));

            newMsg.setDelayTimeLevel(3 + msg.getReconsumeTimes());

            this.mQClientFactory.getDefaultMQProducer().send(newMsg);

        }

    }

从代码中看到DelayTimeLevel =3+reconsumeTime

newMsg.setDelayTimeLevel(3 + msg.getReconsumeTimes());

所以默认重试时，实际是从3开始的，从时间的角度，也验证为什么会重试4次，而且每次间隔的时间是10s/30s/1m .

• 此外发送时，延时等级设置的是9，那么5m之后消费者才能从broker取到该消息消费。消费失败或者异常时，消费者返回给broker，ConsumeConcurrentlyStatus.RECONSUME_LATER，所以broker会重发该消息，由于消费者设置了再次消费的延时级别为5，所以1m之后，消费者才能从broker取出该重发消息。由于consumer没有设置重试次数，所以，broker默认重发该消息16次，然后放入删除队列。

• DefaultRocketMQListenerContainer实现了MessageListenerConcurrently方法，它会循环调用rocketMQListener.onMessage，出现异常会设置delayLevelWhenNextConsume，然后立即返回ConsumeConcurrentlyStatus.RECONSUME_LATER