利用jmx_exporter获取kafka metrics数据

1.下载kafka，链接 https://kafka.apache.org/downloads;
2.配置kafka的JMX PORT,修改bin/kafka-server-start.sh,添加一行export JMX_PORT="9999"，我这里用了9999端口。
3.下载jmx_exporter,链接 https://github.com/prometheus/jmx_exporter, 我将文件放在/opt/kafka/kafka_2.11-1.0.0/jmx_prometheus_javaagent-0.9.jar;
4.创建文件kafka-agent.yaml, 我放在/opt/kafka/kafka_2.11-1.0.0, 内容为你需要的kafka metrics

hostPort: 127.0.0.1:9999 # 这里9999为设置的jmx端口
lowercaseOutputName: true
whitelistObjectNames: 
  - "kafka.controller:type=KafkaController,name=OfflinePartitionsCount"
  - "kafka.controller:type=KafkaController,name=ActiveControllerCount"
  - ......

5.再次修改bin/kafka-server-start.sh,
添加 export KAFKA_OPTS="-javaagent:/opt/kafka/kafka_2.11-1.0.0/jmx_prometheus_javaagent-0.9.jar=9990:/opt/kafka
/kafka_2.11-1.0.0/kafka-agent.yaml";

注意jmx_prometheus_javaagent-0.9.jar=9990:, 9990为jmx_exporter提供数据的端口，之后我们访问http://xxx.xxx.xxx:9990/metrics就能获得需要的metrics数据:

kafka_server_brokertopicmetrics_fiveminuterate{name="TotalFetchRequestsPerSec",} 79706.84209535549
kafka_server_brokertopicmetrics_fiveminuterate{name="TotalProduceRequestsPerSec",} 214.84898012003208
...

6.如果使用了prometheus来抓取数据，只要增加prometheus抓取目标源如下, 再结合grafana来做监控看板。

      - job_name: kafka-jmx
        static_configs:
          - targets:
            - xxx.com:9990
            - xxx:9990
            - xxx:9990