解决Kafka消息堆积及分区不均匀的问题

我在环境中发现代码里面的kafka有所延迟,查看kafka消息发现堆积严重,经过检查发现是kafka消息分区不均匀造成的,消费速度过慢。这里由自己在虚拟机上演示相关问题,给大家提供相应问题的参考思路。

这篇文章有点遗憾并没重现分区不均衡的样例和Warning: Consumer group ‘testGroup1’ is rebalancing. 这里仅将正确的方式展示,等后续重现了在进行补充。

主要有两个要点:

1、一个消费者组只消费一个topic.

2、factory.setConcurrency(concurrency);这里设置监听并发数为 部署单元节点*concurrency=分区数量

1、先在kafka消息中创建对应分区数目的topic(testTopic2,testTopic3)testTopic1由代码创建

./kafka-topics.sh --create --zookeeper 192.168.25.128:2181 --replication-factor 1 --partitions 2 --topic testTopic2

2、添加配置文件application.properties

kafka.test.topic1=testTopic1
kafka.test.topic2=testTopic2
kafka.test.topic3=testTopic3
kafka.broker=192.168.25.128:9092
auto.commit.interval.time=60000
#kafka.test.group=customer-test
kafka.test.group1=testGroup1
kafka.test.group2=testGroup2
kafka.test.group3=testGroup3
kafka.offset=earliest
kafka.auto.commit=false

session.timeout.time=10000
kafka.concurrency=2

3、创建kafka工厂

package com.yin.customer.config;

import org.apache.kafka.clients.consumer.ConsumerConfig;
import org.apache.kafka.common.serialization.StringDeserializer;
import org.springframework.beans.factory.annotation.Value;
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.listener.AbstractMessageListenerContainer;
import org.springframework.kafka.listener.ConcurrentMessageListenerContainer;
import org.springframework.kafka.listener.ContainerProperties;
import org.springframework.stereotype.Component;

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

/**
 * @author yin
 * @Date 2019/11/24 15:54
 * @Method
 */
@Configuration
@Component
public class KafkaConfig {
    @Value("${kafka.broker}")
    private String broker;
    @Value("${kafka.auto.commit}")
    private String autoCommit;

   // @Value("${kafka.test.group}")
    //private String testGroup;

    @Value("${session.timeout.time}")
    private String sessionOutTime;

    @Value("${auto.commit.interval.time}")
    private String autoCommitTime;

    @Value("${kafka.offset}")
    private String offset;
    @Value("${kafka.concurrency}")
    private Integer concurrency;


   @Bean
    KafkaListenerContainerFactory<ConcurrentMessageListenerContainer<String, String>> kafkaListenerContainerFactory(){
        ConcurrentKafkaListenerContainerFactory<String, String> factory = new ConcurrentKafkaListenerContainerFactory<>();
        factory.setConsumerFactory(consumerFactory());
        //监听设置两个个分区
        factory.setConcurrency(concurrency);
        //打开批量拉取数据
        factory.setBatchListener(true);
        //这里设置的是心跳时间也是拉的时间,也就说每间隔max.poll.interval.ms我们就调用一次poll,kafka默认是300s,心跳只能在poll的时候发出,如果连续两次poll的时候超过
        //max.poll.interval.ms 值就会导致rebalance
        //心跳导致GroupCoordinator以为本地consumer节点挂掉了,引发了partition在consumerGroup里的rebalance。
        // 当rebalance后,之前该consumer拥有的分区和offset信息就失效了,同时导致不断的报auto offset commit failed。
        factory.getContainerProperties().setPollTimeout(3000);
        factory.getContainerProperties().setAckMode(ContainerProperties.AckMode.MANUAL_IMMEDIATE);
        return factory;
    }

    private ConsumerFactory<String,String> consumerFactory() {
        return new DefaultKafkaConsumerFactory<String, String>(consumerConfigs());
    }



   @Bean
    public Map<String, Object> consumerConfigs() {
        Map<String, Object> propsMap = new HashMap<>();
        //kafka的地址
        propsMap.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG, broker);
        //是否自动提交 Offset
        propsMap.put(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, autoCommit);
        // enable.auto.commit 设置成 false,那么 auto.commit.interval.ms 也就不被再考虑
        //默认5秒钟,一个 Consumer 将会提交它的 Offset 给 Kafka
        propsMap.put(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG,  5000);

        //这个值必须设置在broker configuration中的group.min.session.timeout.ms 与 group.max.session.timeout.ms之间。
        //zookeeper.session.timeout.ms 默认值:6000
        //ZooKeeper的session的超时时间,如果在这段时间内没有收到ZK的心跳,则会被认为该Kafka server挂掉了。
        // 如果把这个值设置得过低可能被误认为挂掉,如果设置得过高,如果真的挂了,则需要很长时间才能被server得知。
        propsMap.put(ConsumerConfig.SESSION_TIMEOUT_MS_CONFIG, sessionOutTime);
        propsMap.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class);
        propsMap.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class);
        //组与组间的消费者是没有关系的。
        //topic中已有分组消费数据,新建其他分组ID的消费者时,之前分组提交的offset对新建的分组消费不起作用。
        //propsMap.put(ConsumerConfig.GROUP_ID_CONFIG, testGroup);

        //当创建一个新分组的消费者时,auto.offset.reset值为latest时,
        // 表示消费新的数据(从consumer创建开始,后生产的数据),之前产生的数据不消费。
        // https://blog.csdn.net/u012129558/article/details/80427016

        //earliest 当分区下有已提交的offset时,从提交的offset开始消费;无提交的offset时,从头开始消费。
       // latest 当分区下有已提交的offset时,从提交的offset开始消费;无提交的offset时,消费新产生的该分区下的数据。

        propsMap.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG, offset);
        //不是指每次都拉50条数据,而是一次最多拉50条数据()
        propsMap.put(ConsumerConfig.MAX_POLL_RECORDS_CONFIG, 5);
        return propsMap;
    }
}

3、展示kafka消费者

@Component
public class KafkaConsumer {
    private static final Logger logger = LoggerFactory.getLogger(KafkaConsumer.class);


    @KafkaListener(topics = "${kafka.test.topic1}",groupId = "${kafka.test.group1}",containerFactory = "kafkaListenerContainerFactory")
    public void listenPartition1(List<ConsumerRecord<?, ?>> records,Acknowledgment ack) {
        logger.info("testTopic1 recevice a message size :{}" , records.size());


        try {
            for (ConsumerRecord<?, ?> record : records) {
                Optional<?> kafkaMessage = Optional.ofNullable(record.value());
                logger.info("received:{} " , record);
                if (kafkaMessage.isPresent()) {
                    Object message = record.value();
                    String topic = record.topic();
                    Thread.sleep(300);
                    logger.info("p1 topic is:{} received message={}",topic, message);
                }
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            ack.acknowledge();
        }

    }

    @KafkaListener(topics = "${kafka.test.topic2}",groupId = "${kafka.test.group2}",containerFactory = "kafkaListenerContainerFactory")
    public void listenPartition2(List<ConsumerRecord<?, ?>> records,Acknowledgment ack) {
        logger.info("testTopic2 recevice a message size :{}" , records.size());


        try {
            for (ConsumerRecord<?, ?> record : records) {
                Optional<?> kafkaMessage = Optional.ofNullable(record.value());
                logger.info("received:{} " , record);
                if (kafkaMessage.isPresent()) {
                    Object message = record.value();
                    String topic = record.topic();
                    Thread.sleep(300);
                    logger.info("p2 topic :{},received message={}",topic, message);
                }
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            ack.acknowledge();
        }

    }

    @KafkaListener(topics = "${kafka.test.topic3}",groupId = "${kafka.test.group3}",containerFactory = "kafkaListenerContainerFactory")
    public void listenPartition3(List<ConsumerRecord<?, ?>> records, Acknowledgment ack) {
        logger.info("testTopic3 recevice a message size :{}" , records.size());


        try {
            for (ConsumerRecord<?, ?> record : records) {
                Optional<?> kafkaMessage = Optional.ofNullable(record.value());
                logger.info("received:{} " , record);
                if (kafkaMessage.isPresent()) {
                    Object message = record.value();
                    String topic = record.topic();
                    logger.info("p3 topic :{},received message={}",topic, message);
                    Thread.sleep(300);
                }
            }
        } catch (Exception e) {
            e.printStackTrace();
        } finally {
            ack.acknowledge();
        }
    }

}

查看分区消费情况:

7640F867-CBFF-0D77-70AD-1C5AB141E8F9.png

收藏 (0)
评论列表
正在载入评论列表...
我是有底线的
为您推荐
    暂时没有数据