Spring-Kafka 3.0 消费者消费失败处理方案

一、背景

我们作为Kafka在使用Kafka是，必然考虑消息消费失败的重试次数，重试后仍然失败如何处理，要么阻塞，要么丢弃，或者保存

二、设置消费失败重试次数

1 默认重试次数在哪里看

Kafka3.0 版本默认失败重试次数为10次，准确讲应该是1次正常调用+9次重试，这个在这个类可以看到 org.springframework.kafka.listener.SeekUtils

2 如何修改重试次数

据我的实验，spring-kafka3.0版本通过application.yml 配置是行不通的，也没有找到任何一项配置可以改重试次数的（网上很多说的通过配置spring.kafka.consumer.retries 可以配置，我尝试过了，至少3.0版本是不行的，如果有人成功试过可以通过application.yml 配置消费者的消费的重试次数可以留言通知我，谢谢）

经过我不懈努力和尝试，只能通过Java代码配置的方式才可以，并且这种方式相对于application.yml配置更加灵活细致，上代码

    public CommonErrorHandler commonErrorHandler() {
        BackOff backOff = new FixedBackOff(5000L, 3L);
        return  new DefaultErrorHandler(backOff);
    }

然后把这个handler 添加到ConcurrentKafkaListenerContainerFactory中就行了

三、设置消费失败处理方式

1 保存到数据库重试

我们需要在创建DefaultErrorHandler类时加入一个ConsumerAwareRecordRecoverer参数就可以了，这样在重试3次后仍然失败就会保存到数据库中，注意这里save to db成功之后，我认为没有必要执行consumer.commitSync方法，首先这个consumer.commitSync这个方法默认是提交当前批次的最大的offset（可能会导致丢失消息），其次不提交Kafka的消费者仍然回去消费后面的消息，只要后面的消息，消费成功了，那么依然会提交offset，覆盖了这个offset

    public CommonErrorHandler commonErrorHandler() {
        // 创建 FixedBackOff 对象
        BackOff backOff = new FixedBackOff(5000L, 3L);
        DefaultErrorHandler defaultErrorHandler = new DefaultErrorHandler((ConsumerAwareRecordRecoverer) (record, consumer, exception) -> {
            log.info("save to db " + record.value().toString());
        }, backOff);
        return defaultErrorHandler;
    }

如果你硬要提交也可以试试下面这种，指定提交当前的offset

    public CommonErrorHandler commonErrorHandler() {
        // 创建 FixedBackOff 对象
        BackOff backOff = new FixedBackOff(5000L, 3L);
        DefaultErrorHandler defaultErrorHandler = new DefaultErrorHandler((ConsumerAwareRecordRecoverer) (record, consumer, exception) -> {
            log.info("save to db " + record.value().toString());
            Map offsets = new HashMap();
            offsets.put(new TopicPartition(record.topic(),record.partition()),new OffsetAndMetadata(record.offset()));
            consumer.commitSync(offsets);
        }, backOff);
        return defaultErrorHandler;
    }

2 发送到Kafka死信队列

仍然在创建DefaultErrorHandler类时加入一个DeadLetterPublishingRecoverer 类就行了，默认会把消息发到kafkaTemplate 配置的topic名字为your_topic+.DLT

@Autowired
private KafkaTemplate kafkaTemplate;
   

 public CommonErrorHandler commonErrorHandler() {
        // 创建 FixedBackOff 对象
        BackOff backOff = new FixedBackOff(5000L, 3L);

        DefaultErrorHandler defaultErrorHandler = new DefaultErrorHandler(new DeadLetterPublishingRecoverer(kafkaTemplate), backOff);

        return defaultErrorHandler;
    }

ConsumerRecordRecoverer 接口总共就这2种实现方式

四、整体消费者代码粘贴

1 application.yml

kafka-consumer:
  bootstrapServers: 192.168.31.114:9092
  groupId: goods-center
  #后台的心跳线程必须在30秒之内提交心跳,否则会reBalance
  sessionTimeOut: 30000
  autoOffsetReset: latest
  #取消自动提交,即便如此 spring会帮助我们自动提交
  enableAutoCommit: false
  #自动提交间隔
  autoCommitInterval: 1000
  #拉取的最小字节
  fetchMinSize: 1
  #拉去最小字节的最大等待时间
  fetchMaxWait: 500
  maxPollRecords: 50
  #300秒的提交间隔,如果程序大于300秒提交,会报错
  maxPollInterval: 300000
  #心跳间隔
  heartbeatInterval: 10000
  keyDeserializer: org.apache.kafka.common.serialization.LongDeserializer
  valueDeserializer: org.springframework.kafka.support.serializer.JsonDeserializer

2 KafkaListenerProperties

package com.ychen.goodscenter.fafka;

import lombok.Getter;
import lombok.Setter;
import org.springframework.boot.context.properties.ConfigurationProperties;
import org.springframework.context.annotation.Configuration;

@Configuration
//指定配置文件的前缀
@ConfigurationProperties(prefix = "kafka-consumer")
@Getter
@Setter
public class KafkaListenerProperties {
 
    private String groupId;
 
    private String sessionTimeOut;
 
    private String bootstrapServers;
 
    private String autoOffsetReset;
 
    private boolean enableAutoCommit;
 
    private String autoCommitInterval;
 
    private String fetchMinSize;
 
    private String fetchMaxWait;
 
    private String maxPollRecords;
 
    private String maxPollInterval;
 
    private String heartbeatInterval;
 
    private String keyDeserializer;
 
    private String valueDeserializer;
 
}

3 KafkaConsumerConfig

package com.ychen.goodscenter.fafka;

import com.alibaba.fastjson2.JSONObject;
import lombok.extern.slf4j.Slf4j;
import org.apache.kafka.clients.consumer.ConsumerConfig;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.boot.context.properties.EnableConfigurationProperties;
import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.kafka.config.ConcurrentKafkaListenerContainerFactory;
import org.springframework.kafka.config.KafkaListenerContainerFactory;
import org.springframework.kafka.core.ConsumerFactory;
import org.springframework.kafka.core.DefaultKafkaConsumerFactory;
import org.springframework.kafka.core.KafkaTemplate;
import org.springframework.kafka.listener.*;
import org.springframework.util.backoff.BackOff;
import org.springframework.util.backoff.FixedBackOff;

import java.util.HashMap;
import java.util.Map;

@Configuration
@EnableConfigurationProperties(KafkaListenerProperties.class)
@Slf4j
public class KafkaConsumerConfig {
    @Autowired
    private KafkaListenerProperties kafkaListenerProperties;

    @Autowired
    private KafkaTemplate kafkaTemplate;

    @Bean
    public KafkaListenerContainerFactory> kafkaListenerContainerFactory() {
        ConcurrentKafkaListenerContainerFactory factory = new ConcurrentKafkaListenerContainerFactory();
        factory.setConsumerFactory(consumerFactory());
        // 并发数 多个微服务实例会均分
        factory.setConcurrency(2);
//        factory.setBatchListener(true);
        factory.setCommonErrorHandler(commonErrorHandler());

        ContainerProperties containerProperties = factory.getContainerProperties();
        // 是否设置手动提交
        containerProperties.setAckMode(ContainerProperties.AckMode.MANUAL_IMMEDIATE);

        return factory;
    }


    private ConsumerFactory consumerFactory() {
        Map consumerConfigs = consumerConfigs();
        log.info("消费者的配置信息:{}", JSONObject.toJSONString(consumerConfigs));
        return new DefaultKafkaConsumerFactory(consumerConfigs);
    }


    public CommonErrorHandler commonErrorHandler() {
        // 创建 FixedBackOff 对象
        BackOff backOff = new FixedBackOff(5000L, 3L);

        DefaultErrorHandler defaultErrorHandler = new DefaultErrorHandler(new DeadLetterPublishingRecoverer(kafkaTemplate), backOff);
//        DefaultErrorHandler defaultErrorHandler = new DefaultErrorHandler((ConsumerAwareRecordRecoverer) (record, consumer, exception) -> {
//            log.info("save to db " + record.value().toString());
//            Map offsets = new HashMap();
//            offsets.put(new TopicPartition(record.topic(),record.partition()),new OffsetAndMetadata(record.offset()));
//            consumer.commitSync(offsets);
//        }, backOff);
        return defaultErrorHandler;
    }

    public Map consumerConfigs() {
        Map propsMap = new HashMap();
        // 服务器地址
        propsMap.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, kafkaListenerProperties.getBootstrapServers());
        // 是否自动提交
        propsMap.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, kafkaListenerProperties.isEnableAutoCommit());
        // 自动提交间隔
        propsMap.put(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG, kafkaListenerProperties.getAutoCommitInterval());

        //会话时间
        propsMap.put(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG, kafkaListenerProperties.getSessionTimeOut());
        //key序列化
        propsMap.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, kafkaListenerProperties.getKeyDeserializer());
        //value序列化
        propsMap.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, kafkaListenerProperties.getValueDeserializer());
        // 心跳时间
        propsMap.put(ConsumerConfig.HEARTBEAT_INTERVAL_MS_CONFIG, kafkaListenerProperties.getHeartbeatInterval());

        // 分组id
        propsMap.put(ConsumerConfig.GROUP_ID_CONFIG, kafkaListenerProperties.getGroupId());
        //消费策略
        propsMap.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, kafkaListenerProperties.getAutoOffsetReset());
        // poll记录数
        propsMap.put(ConsumerConfig.MAX_POLL_RECORDS_CONFIG, kafkaListenerProperties.getMaxPollRecords());
        //poll时间
        propsMap.put(ConsumerConfig.MAX_POLL_INTERVAL_MS_CONFIG, kafkaListenerProperties.getMaxPollInterval());

        propsMap.put("spring.json.trusted.packages", "com.ychen.**");

        return propsMap;
    }


}

4 MessageListener

package com.ychen.goodscenter.fafka;

import com.ychen.goodscenter.service.OrderService;
import com.ychen.goodscenter.vo.req.SubmitOrderReq;
import lombok.extern.slf4j.Slf4j;
import org.apache.kafka.clients.consumer.ConsumerRecord;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.dao.DuplicateKeyException;
import org.springframework.kafka.annotation.KafkaListener;
import org.springframework.kafka.support.Acknowledgment;
import org.springframework.stereotype.Component;

@Component
@Slf4j
public class MessageListener {
    @Autowired
    private OrderService orderService;

    @KafkaListener(topics = "order-message-topic", containerFactory = "kafkaListenerContainerFactory")
    public void processMessage(ConsumerRecord record, Acknowledgment acknowledgment) {
        log.info("order-message-topic message Listener, Thread ID: " + Thread.currentThread().getId());

        try {
            log.info("order-message-topic message received, orderId: {}", record.value().getOrderId());

            orderService.submitOrder(record.value());
            // 同步提交
            acknowledgment.acknowledge();
            log.info("order-message-topic message acked: orderId: {}", record.value().getOrderId());
        } catch (DuplicateKeyException dupe) {
            // 处理异常情况
            log.error("order-message-topic message error DuplicateKeyException", dupe);
            // 重复数据，忽略掉，同步提交
            acknowledgment.acknowledge();
        }
    }
}