prometheus监控zookeeper

官方介绍:主要有两种方式,四字命令和jmx

1、四字命令:

https://zookeeper.apache.org/doc/r3.4.14/zookeeperAdmin.html#sc_zkCommands 其中最常用的三个stat、srvr、cons

Three of the more interesting commands: "stat" gives some general information about the server and connected clients, while "srvr" and "cons" give extended details on server and connections respectively.

$ echo srvr | nc svc-zk.o.svc.mhc.local 2181   
Zookeeper version: 3.4.10-39d3a4f269333c922ed3db283be479f9deacaa0f, built on 03/23/2017 10:13 GMT
Latency min/avg/max: 0/0/802
Received: 5833450
Sent: 6074801
Connections: 84
Outstanding: 0
Zxid: 0x2d9407
Mode: standalone
Node count: 12977

2、jmx监控

是利用jvm_exporter暴露指标,参考prometheus官方jmx_exporter https://github.com/prometheus/jmx_exporter

jvm-exporter从基于jvm的应用读取jmx数据并以http的方式暴露给prometheus抓取
jmx是一种常用的技术,输出运行中应用的状态信息,并且可以控制它(例如用jmx触发gc)
jvm-exporter是用jmx apis收集app和jvm metrics的Java应用,以Java agent方式运行在同一个jvm里面。
jvm-exporter用Java写的,以jar包的方式发布,下载地址
有一个ansible role(https://github.com/alexdzyoba/ansible-jmx-exporter),配置文件包含将JMX MBean重写为Prometheus展示格式度量标准的规则。基本上它是将MBeans字符串转换为Prometheus字符串的regexp集合。

zookeeper是很多系统的关键组成部分,例如Hadoop, Kafka and Clickhouse,因此监控是必不可少的。尽管可以使用4字命令(mntr,stat等)输出状态信息,但我更喜欢使用JMX,以避免常见的Zookeeper查询(它们会向metrics添加噪声干扰,4字命令被视为正常的Zookeeper请求)

2.1、以java-agent方式运行(官方推荐):

zookeeper启动的时候传入参数启用,不需要开启jmx

$ cat java.env 
export SERVER_JVMFLAGS="-javaagent:/root/jmx_prometheus_javaagent-0.12.0.jar=7070:/root/config.yaml $SERVER_JVMFLAGS"

查看启动日志并检查结果:

$ netstat -tlnp | grep 7070
tcp        0      0 0.0.0.0:7070            0.0.0.0:*               LISTEN      892/java

$ curl -s localhost:7070/metrics | head
# HELP jvm_threads_current Current thread count of a JVM
# TYPE jvm_threads_current gauge
jvm_threads_current 16.0
# HELP jvm_threads_daemon Daemon thread count of a JVM
# TYPE jvm_threads_daemon gauge
jvm_threads_daemon 12.0
# HELP jvm_threads_peak Peak thread count of a JVM
# TYPE jvm_threads_peak gauge
jvm_threads_peak 16.0
# HELP jvm_threads_started_total Started thread count of a JVM

With jmx-exporter you can scrape the metrics of running JVM applications. jmx-exporter runs as a Java agent (inside the target JVM) scrapes JMX metrics, rewrite it according to config rules and exposes it in Prometheus exposition format.

容器中启用jvm_exporter

基于官方zookeeper的github仓库,修改官方dockerfile以加入启动参数
下面链接是我修改后的,亲测可用

https://github.com/weifan01/zookeeper-docker/tree/master/3.4.14

构建好的zk-3.4.14版本镜像: docker pull 3070656869/zookeeper:3.4.14-jvm-exporter
启动OK后,访问8080端口查看结果

2.2、以http方式独立运行

参考官方文档:https://github.com/prometheus/jmx_exporter

git clone https://github.com/prometheus/jmx_exporter.git
cd jmx_exporter
mvn package
#修改配置文件example_configs/httpserver_sample_config.yml
#修改启动脚本里面的监听地址和端口
sh run_sample_httpserver.sh

我测试配置如下:

$ cat http_server_config.yaml 
---
# jmx地址和端口
hostPort: 172.21.10.248:6666
username: 
password: 

rules:
  # replicated Zookeeper
  - pattern: "org.apache.ZooKeeperService<name0=ReplicatedServer_id(\\d+)><>(\\w+)"
    name: "zookeeper_$2"
  - pattern: "org.apache.ZooKeeperService<name0=ReplicatedServer_id(\\d+), name1=replica.(\\d+)><>(\\w+)"
    name: "zookeeper_$3"
    labels:
      replicaId: "$2"
  - pattern: "org.apache.ZooKeeperService<name0=ReplicatedServer_id(\\d+), name1=replica.(\\d+), name2=(\\w+)><>(\\w+)"
    name: "zookeeper_$4"
    labels:
      replicaId: "$2"
      memberType: "$3"
  - pattern: "org.apache.ZooKeeperService<name0=ReplicatedServer_id(\\d+), name1=replica.(\\d+), name2=(\\w+), name3=(\\w+)><>(\\w+)"
    name: "zookeeper_$4_$5"
    labels:
      replicaId: "$2"
      memberType: "$3"
  # standalone Zookeeper
  - pattern: "org.apache.ZooKeeperService<name0=StandaloneServer_port(\\d+)><>(\\w+)"
    name: "zookeeper_$2"
  - pattern: "org.apache.ZooKeeperService<name0=StandaloneServer_port(\\d+), name1=InMemoryDataTree><>(\\w+)"
    name: "zookeeper_$2"

启动:

java -Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote.port=6667 -jar jmx_prometheus_httpserver-0.12.1-SNAPSHOT-jar-with-dependencies.jar 172.21.10.248:6668 http_server_config.yaml

访问:
利用prometheus监控zookeeper

prometheus监控

在kubernetes集群中,抓取集群中每个zk的指标,通过service访问不清楚是具体哪个zk节点上的数据,因此需要根据pod的域名去访问,statefulset类型的pod都有一个dns记录,访问pod的域名格式为:
podname.headless-name.namespace.svc.mhc.local

prometheus配置中添加:

    - job_name: zk-s
      static_configs:
      - targets:
        - stateful-zk-0.svc-zk.s.svc.mhc.local:8081
        - stateful-zk-1.svc-zk.s.svc.mhc.local:8081
        - stateful-zk-2.svc-zk.s.svc.mhc.local:8081

发送reload信号给prometheus
在prometheus控制台target中可以看到已经有任务up了

如有误,请指正,谢谢!

Logo

权威|前沿|技术|干货|国内首个API全生命周期开发者社区

更多推荐