*本文基于SpringBoot整合Kafka，通过简单配置实现生产及消费，包括生产消费的配置说明、消费者偏移设置方式等。更多功能细节可参考

spring kafka 文档：https://docs.spring.io/spring-kafka/docs/current/reference/html

前提条件

• 搭建Kafka环境，参考Kafka集群环境搭建及使用
• Java环境：JDK1.8
• Maven版本：apache-maven-3.6.3
• 开发工具：IntelliJ IDEA

项目环境

1. 创建Springboot项目。
2. pom.xml文件中引入kafka依赖。

<dependencies>
        <dependency>
            <groupId>org.springframework.kafka</groupId>
            <artifactId>spring-kafka</artifactId>
        </dependency>
</dependencies>

创建Topic

创建topic命名为testtopic并指定2个分区。

./kafka-topics.sh --bootstrap-server 127.0.0.1:9092 --create --topic testtopic --partitions 2

配置信息

application.yml配置文件信息

spring:
  application:
    name: kafka_springboot
  kafka:
    bootstrap-servers: 127.0.0.1:9092
    producer:
      #ACK机制，默认为1 (0,1,-1)
      acks: -1
      key-serializer: org.apache.kafka.common.serialization.StringSerializer
      value-serializer: org.apache.kafka.common.serialization.StringSerializer
      properties:
        # 自定义分区策略
        partitioner:
          class: org.bg.kafka.PartitionPolicy

    consumer:
      #设置是否自动提交，默认为true
      enable-auto-commit: false
      key-deserializer: org.apache.kafka.common.serialization.StringDeserializer
      value-deserializer: org.apache.kafka.common.serialization.StringDeserializer
      #当一个新的消费组或者消费信息丢失后，在哪里开始进行消费。earliest：消费最早的消息。latest（默认）：消费最近可用的消息。none：没有找到消费组消费数据时报异常。
      auto-offset-reset: latest
      #批量消费时每次poll的数量
      #max-poll-records: 5
    listener:
      #      当每一条记录被消费者监听器处理之后提交
      #      RECORD,
      #      当每一批数据被消费者监听器处理之后提交
      #      BATCH,
      #      当每一批数据被消费者监听器处理之后，距离上次提交时间大于TIME时提交
      #      TIME,
      #      当每一批数据被消费者监听器处理之后，被处理record数量大于等于COUNT时提交
      #      COUNT,
      #      #TIME |　COUNT　有一个条件满足时提交
      #      COUNT_TIME,
      #      #当每一批数据被消费者监听器处理之后,手动调用Acknowledgment.acknowledge()后提交:
      #      MANUAL,
      #      # 手动调用Acknowledgment.acknowledge()后立即提交
      #      MANUAL_IMMEDIATE;
      ack-mode: manual
      #批量消费
      type: batch

更多配置信息查看KafkaProperties

生产消息

@Component
public class Producer {
    @Autowired
    private KafkaTemplate kafkaTemplate;

    public void send(String msg) {
        kafkaTemplate.send(new ProducerRecord<String, String>("testtopic", "key111", msg));
    }
}

生产自定义分区策略

package org.bg.kafka;

import org.apache.kafka.clients.producer.Partitioner;
import org.apache.kafka.common.Cluster;
import org.apache.kafka.common.PartitionInfo;
import org.apache.kafka.common.utils.Utils;

import java.util.List;
import java.util.Map;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.ThreadLocalRandom;
import java.util.concurrent.atomic.AtomicInteger;

public class PartitionPolicy implements Partitioner {

    private final ConcurrentMap<String, AtomicInteger> topicCounterMap = new ConcurrentHashMap();

    @Override
    public int partition(String topic, Object key, byte[] keyBytes, Object value, byte[] valueBytes, Cluster cluster) {
        List<PartitionInfo> partitions = cluster.partitionsForTopic(topic);
        int numPartitions = partitions.size();
        if (keyBytes == null) {
            int nextValue = this.nextValue(topic);
            List<PartitionInfo> availablePartitions = cluster.availablePartitionsForTopic(topic);
            if (availablePartitions.size() > 0) {
                int part = Utils.toPositive(nextValue) % availablePartitions.size();
                return ((PartitionInfo)availablePartitions.get(part)).partition();
            } else {
                return Utils.toPositive(nextValue) % numPartitions;
            }
        } else {
            return Utils.toPositive(Utils.murmur2(keyBytes)) % numPartitions;
        }
    }


    private int nextValue(String topic) {
        AtomicInteger counter = (AtomicInteger)this.topicCounterMap.get(topic);
        if (null == counter) {
            counter = new AtomicInteger(ThreadLocalRandom.current().nextInt());
            AtomicInteger currentCounter = (AtomicInteger)this.topicCounterMap.putIfAbsent(topic, counter);
            if (currentCounter != null) {
                counter = currentCounter;
            }
        }

        return counter.getAndIncrement();
    }

    @Override
    public void close() {

    }

    @Override
    public void configure(Map<String, ?> map) {

    }
}

生产到指定分区

ProducerRecord有指定分区的构造方法，设置分区号
public ProducerRecord(String topic, Integer partition, K key, V value)。

kafkaTemplate.send(new ProducerRecord<String, String>("testtopic",1, "key111", msg));

消费消息

/**
 * 自定义seek参考
 * https://docs.spring.io/spring-kafka/docs/current/reference/html/#seek
 */
@Component
public class Consumer implements ConsumerSeekAware{


    @KafkaListener(topics = {"testtopic"},groupId = "test_group",clientIdPrefix = "bg",id = "testconsumer")
    public void onMessage(List<ConsumerRecord<String, String>> records, Acknowledgment ack){
        System.out.println(records.size());
        System.out.println(records.toString());
        ack.acknowledge();
    }


    @Override
    public void onPartitionsAssigned(Map<TopicPartition, Long> assignments, ConsumerSeekCallback callback) {
        //按照时间戳设置偏移
        callback.seekToTimestamp(assignments.keySet(),1670233826705L);
        //设置偏移到最近
        callback.seekToEnd(assignments.keySet());
        //设置偏移到最开始
        callback.seekToBeginning(assignments.keySet());
        //指定 offset
        for (TopicPartition topicPartition : assignments.keySet()) {
            callback.seek(topicPartition.topic(),topicPartition.partition(),0L);
        }

    }

}

offset设置方式

如代码所示，实现ConsumerSeekAware接口，设置offset几种方式：

• 指定 offset，需要自己维护 offset，方便重试。
• 指定从头开始消费。
• 指定 offset 为最近可用的 offset (默认)。
• 根据时间戳获取 offset，设置 offset。

代码仓库

https://gitee.com/codeWBG/learn_kafka