前言

对于 Kafka 组件而言，我们通常会对 kafka 服务端添加一些监控，来确保服务的稳定性，虽然有 kafka-exporter 来对消费者进行监控，但是指标很少，对于生产者和消费者更细粒度的监控就无法做到了。只能将监控部署在客户端上，这样我们就能拿到更加详细的监控数据。

例如：对于消费者来说，我们可以获取到消费者重平衡次数，每次poll 数据的条数等等。

这里梳理下如何对消费者添加 JMX 监控。

一、所需组件

这里我们只针对于 JAVA 语言的 KAFKA 客户端为例，需要以下东西：

• kafka-client:3.4.1
• JMX-exporter

需要提前准备好 kafka 服务和创建好topic。

二、编写测试代码

1. pom 文件

 <properties>
        <maven.compiler.source>8</maven.compiler.source>
        <maven.compiler.target>8</maven.compiler.target>
        <kafka.version>3.4.1</kafka.version>
    </properties>

    <dependencies>
        <dependency>
            <groupId>org.apache.kafka</groupId>
            <artifactId>kafka-clients</artifactId>
            <version>${kafka.version}</version>
        </dependency>
    </dependencies>


    <build>
        <plugins>
            <plugin>
                <groupId>org.apache.maven.plugins</groupId>
                <artifactId>maven-assembly-plugin</artifactId>
                <version>3.3.0</version>
                <configuration>
                    <descriptorRefs>
                        <descriptorRef>jar-with-dependencies</descriptorRef>
                    </descriptorRefs>
                </configuration>
                <executions>
                    <execution>
                        <id>make-assembly</id>
                        <phase>package</phase>
                        <goals>
                            <goal>single</goal>
                        </goals>
                    </execution>
                </executions>
            </plugin>
        </plugins>
    </build>

2. 生产者代码

public class ProducerDemo {

    public static void main(String[] args) {
        //主题（当主题不存在，自动创建主题）
        String topic = "test";
        //配置
        Properties properties = new Properties();
        //kafka服务器地址
        properties.put(ProducerConfig.BOOTSTRAP_SERVERS_CONFIG,"192.168.195.132:9092");
        //反序列化器
        properties.put(ProducerConfig.KEY_SERIALIZER_CLASS_CONFIG, StringSerializer.class);
        properties.put(ProducerConfig.VALUE_SERIALIZER_CLASS_CONFIG,StringSerializer.class);

        //生产者
        KafkaProducer<String,String> kafkaProducer = new KafkaProducer(properties);

        //生产信息
        for (int i = 0; i < 100; i++) {
            String msg = String.format("hello,第%d条信息", i);
            //消息（key可以为null,key值影响消息发往哪个分区）
            ProducerRecord<String, String> producerRecord = new ProducerRecord<String, String>(topic, String.valueOf(i), msg);
            //发送
            kafkaProducer.send(producerRecord);
            System.out.println("发送第"+i+"条信息");
        }
        //关闭
        kafkaProducer.close();
    }
}

3. 消费者代码

public class ConsumerDemo {

    public static void main(String[] args) throws Exception{
        //主题
        String topic = "test";

        //配置
        Properties properties = new Properties();
        //kafka服务器地址
        properties.put(ConsumerConfig.BOOTSTRAP_SERVERS_CONFIG,"192.168.195.132:9092");
        //k，v的序列化器
        properties.put(ConsumerConfig.KEY_DESERIALIZER_CLASS_CONFIG, StringDeserializer.class);
        properties.put(ConsumerConfig.VALUE_DESERIALIZER_CLASS_CONFIG,StringDeserializer.class);
        //消费者分组
        properties.put(ConsumerConfig.GROUP_ID_CONFIG,"Consumer-Group-1");
        //offset重置模式
        properties.put(ConsumerConfig.AUTO_OFFSET_RESET_CONFIG,"earliest");

        //消费者
        KafkaConsumer<String, String> kafkaConsumer = new KafkaConsumer(properties);
        //订阅（可以订阅多个主题）
        kafkaConsumer.subscribe(Collections.singletonList(topic));

        //消费
        while (true){
            //获取信息
            ConsumerRecords<String, String> records = kafkaConsumer.poll(Duration.ofMillis(1000));
            //遍历
            records.forEach(o->{
                System.out.println(String.format("topic==%s,offset==%s,key==%s,value==%s",o.topic(),o.offset(),o.key(),o.value()));
            });
            //睡眠
            Thread.sleep(500);
        }
    }
}

4. 打包

mvn clean package

5. 测试生产和消费

生产数据测试：

java -cp kafka-1.0-jar-with-dependencies.jar      ProducerDemo

消费数据测试：

java -cp kafka-1.0-jar-with-dependencies.jar      ConsumerDemo

三、使用jmx-agent

1. 下载

https://repo1.maven.org/maven2/io/prometheus/jmx/jmx_prometheus_javaagent/0.18.0/jmx_prometheus_javaagent-0.18.0.jar

2. 编写kafka-client.yaml

---
lowercaseOutputName: true
lowercaseOutputLabelNames: true
whitelistObjectNames:
  - "kafka.consumer:*"
  - "kafka.producer:*"
blacklistObjectNames:
  - "kafka.admin.client:*"
  - "kafka.consumer:type=*,id=*"
  - "kafka.producer:type=*,id=*"
  - "kafka.*:type=kafka-metrics-count,*"
rules:
  # "kafka.consumer:type=app-info,client-id=*"
  # "kafka.producer:type=app-info,client-id=*"
  - pattern: "kafka.(.+)<type=app-info, client-id=(.+)><>(.+): (.+)"
    value: 1
    name: kafka_$1_app_info
    cache: true
    labels:
      client_type: $1
      client_id: $2
      $3: $4
    type: UNTYPED
  # "kafka.consumer:type=consumer-metrics,client-id=*, protocol=*, cipher=*"
  # "kafka.consumer:type=type=consumer-fetch-manager-metrics,client-id=*, topic=*, partition=*"
  # "kafka.producer:type=producer-metrics,client-id=*, protocol=*, cipher=*"
  - pattern: "kafka.(.+)<type=(.+), (.+)=(.+), (.+)=(.+), (.+)=(.+)><>(.+):"
    name: kafka_$1_$2_$9
    type: GAUGE
    cache: true
    labels:
      client_type: $1
      $3: "$4"
      $5: "$6"
      $7: "$8"
  # "kafka.consumer:type=consumer-node-metrics,client-id=*, node-id=*"
  # "kafka.consumer:type=consumer-fetch-manager-metrics,client-id=*, topic=*"
  # "kafka.producer:type=producer-node-metrics,client-id=*, node-id=*"
  # "kafka.producer:type=producer-topic-metrics,client-id=*, topic=*"
  - pattern: "kafka.(.+)<type=(.+), (.+)=(.+), (.+)=(.+)><>(.+):"
    name: kafka_$1_$2_$7
    type: GAUGE
    cache: true
    labels:
      client_type: $1
      $3: "$4"
      $5: "$6"
  # "kafka.consumer:type=consumer-fetch-manager-metrics,client-id=*"
  # "kafka.consumer:type=consumer-metrics,client-id=*"
  # "kafka.producer:type=producer-metrics,client-id=*"
  - pattern: "kafka.(.+)<type=(.+), (.+)=(.+)><>(.+):"
    name: kafka_$1_$2_$5
    type: GAUGE
    cache: true
    labels:
      client_type: $1
      $3: "$4"
  - pattern: "kafka.(.+)<type=(.+)><>(.+):"
    name: kafka_$1_$2_$3
    cache: true
    labels:
      client_type: $1

3. 重新启动消费程序

java -cp kafka-1.0-jar-with-dependencies.jar -javaagent:/opt/javaProject/jmx_prometheus_javaagent-0.18.0.jar=1234:/opt/javaProject/kafka_client.yml ConsumerDemo

4. 打开web ui

访问http://xxx.xxx.xxx.xxx:1234

这里就能看到我们采集到了kafka 消费者实例的详细指标。我们就可以和Prometheus进行集成，这里就不详细介绍了。

四、具体指标

想要获取更多的指标和指标具体的用途可以参考官网的文档
https://kafka.apache.org/documentation/#consumer_monitoring

具体指标如下：
以下是添加了中文翻译后的Kafka Consumer Metrics表格：

Consumer Coordinator Metrics

Consumer Fetch Metrics