kafka监控配置和告警配置——筑梦之路

kafka_exporter项目地址:https://github.com/danielqsj/kafka_exporter

docker-compose部署kafka_exporter

bash 复制代码
# docker-compose部署多个kafka_exporter,每个exporter对接一个kafka

# cat docker-compose.yml

version: '3.1'
services:
  kafka-exporter-opslogs:
    image: bitnami/kafka-exporter:latest
    command:
      - '--kafka.server=10.2.19.43:9092'
      - '--kafka.server=10.2.24.62:9092'
      - '--kafka.server=10.5.98.190:9092'
      - '--kafka.version=3.2.1'
    restart: always
    ports:
      - 9310:9308

  kafka-exporter-prod:
    image: bitnami/kafka-exporter:latest
    command:
      - '--kafka.server=192.168.53.99:9092'
      - '--kafka.server=192.168.53.53:9092'
      - '--kafka.server=192.168.53.96:9092'
    restart: always
    ports:
      - 9311:9308

注意:配置上每个kafka broker的地址,kafka3需要指定版本

Promethus配置job接入kafka-exporter

bash 复制代码
- job_name: 'kafka-exporter'
    metrics_path: /metrics
    scrape_interval: 15s
    scrape_timeout: 10s
    static_configs:
    - targets:
      - 10.0.0.26:9310
      labels:
        name: kafka-opslogs
    - targets:
      - 10.0.0.26:9311
      labels:
        name: kafka-prod

注意:每个kafka-exporter必须增加 name标签,看板需要使用这个标签

KAFKA Grafana Dashboard

全局信息、消费者与Topic、异常与积压分析

分区维度明细

Prometheus告警规则

bash 复制代码
- name: kafka
  rules:
  - alert: KAFKA_brokers异常
    expr: kafka_broker_info != 1
    for: 2m
    labels:
      severity: critical
    annotations:
      description: "{{ $labels.name }}当前brokers异常:{{ $labels.address }}"

  - alert: 电商生产KAFKA消息整体积压
    expr: sum(kafka_consumergroup_lag_sum{job="kafka-exporter"}) by (name,consumergroup, topic)>5000
    for: 2m
    labels:
      severity: critical
    annotations:
      description: "【环境】{{ $labels.name }}\n【消费组】{{ $labels.consumergroup }}\n【topic】{{ $labels.topic }}【积压】:{{ $value | printf \"%.2f\" }}"

  - alert: 电商生产KAFKA消息分区积压
    expr: (sum(kafka_consumergroup_lag{job="kafka-exporter"}) by (name,consumergroup, topic, partition)>1500) AND ON() (hour()+8)%24 >= 7 <= 21
    for: 3m
    labels:
      severity: critical
    annotations:
      description: "【环境】{{ $labels.name }}\n【消费组】{{ $labels.consumergroup }}\n【topic】{{$labels.topic}}【分区】{{ $labels.partition }}【积压】:{{ $value | printf \"%.2f\" }}"

  - alert: 电商生产KAFKA分区数过多
    expr: sum by(name)(kafka_topic_partitions{job="kafka-exporter",topic !~"__.*"})>1500
    for: 2m
    labels:
      severity: critical
    annotations:
      description: "{{ $labels.name }}当前分区数:{{ $value | printf \"%.2f\" }}"

  - alert: 电商生产KAFKA_brokers丢失
    expr: kafka_brokers{job="kafka-exporter"} < 3
    for: 2m
    labels:
      severity: critical
    annotations:
      description: "{{ $labels.name }}当前brokers数:{{ $value | printf \"%.2f\" }}"

  - alert: 电商生产KAFKA_TopicsReplicas
    expr: sum(kafka_topic_partition_in_sync_replica{job="kafka-exporter"}) by (name,topic) <1
    for: 2m
    labels:
      severity: critical
    annotations:
      description: "{{ $labels.name }} Kafka topic in-sync partition:{{ $value | printf \"%.2f\" }}"
相关推荐
用户531397318173 小时前
[Kafka源码揭秘] 消息写入全链路:网络IO与请求入队
kafka·消息队列
swordbob5 小时前
一个 Partition就是一个年级,一个消费者组占用1年级1班,另一个消费者组占用1年级2班,座位号(offset)可以相同,但班级号不同?
kafka
香吧香6 小时前
Kafka 三节点集群:只订阅一个 broker 会丢消息吗?能用 VIP订阅 吗?
kafka
盟道科技7 小时前
小程序订阅消息大规模投递实践:频控令牌池、Redis限流与失败补偿设计
分布式·算法·小程序
雾隐隐o7 小时前
Kafka 生产者消息丢失:ACK、重试与可靠发送
kafka
卷毛的技术笔记8 小时前
RocketMQ事务消息:我把分布式事务这层窗户纸捅破了
java·分布式·后端·java-rocketmq
heimeiyingwang9 小时前
【中台·技术篇】消息队列中台:Kafka/RocketMQ 统一管理与多租户隔离
kafka·rocketmq·中台
程序员天天困1 天前
Kafka 接入 AI 的三条路线:MCP 提案、会话记忆与实时上下文
大数据·后端·kafka
懂软件的胡子个哥1 天前
微信工单系统如何基于 WechatApi 做消息分流和状态流转
运维·分布式·微信·架构·企业微信
九皇叔叔2 天前
从本地事务到分布式事务
java·分布式·分布式事务·cap·base