《Kubernetes监控篇：Kubernetes单节点之微服务JVM内存监控》

痛定思痛。 2022-08-29 15:49 429阅读 0赞

### 文章目录 ###

*  一、背景信息
 *  二、监控方案
 *  三、环境信息
 *  四、部署操作
 *   *   *  4.1、资源下载
         *  4.2、修改Dockerfile
         *  4.3、修改mpmt-adapter.yaml文件
         *  4.4、部署prometheus
         *  4.5、导入dashborard模板
 *  总结：整理不易，如果对你有帮助，可否点赞关注一下？

--------------------

# 一、背景信息 #

**复杂和高并发下的服务，必须保证每次gc不会出现性能下降，各种性能指标不会出现波动，gc回收规律而且干净，所以必须找到合适的jvm设置。而要找到合适的jvm设置，必须要了解当前java服务的jvm使用情况。所以我觉得非常有必要对java服务的jvm使用情况实时监控。**

**业务系统涉及到的java服务分为：tomcat类型、jar包类型、k8s微服务类型。**

--------------------

# 二、监控方案 #

**方案一：业务系统java服务采用的是SpringBoot框架，SpringBoot自带监控功能Actuator，可以帮助实现对程序内部运行情况监控，比如监控状况、Bean加载情况、环境变量、日志信息、线程信息、健康检查、审计、统计和HTTP追踪等。Actuator同时还可以与外部应用监控系统整合，比如Prometheus，可以选择使用HTTP端点或JMX来管理和监视应用程序。**

**方案二：通过JMX Exporter来暴露 Java 应用的JVM监控指标，JMX Exporter有两种用法。**

# 方法一：启动独立进程
    JVM启动时指定参数，暴露JMX的RMI接口，JMX-Exporter调用RMI获取JVM运行时状态数据，
    转换为Prometheus metrics格式，并暴露端口让Prometheus采集。
    
    # 方法二：JVM进程内启动(in-process)
    JVM启动时指定参数，通过javaagent 的形式运行JMX-Exporter 的jar包，
    进程内读取JVM运行时状态数据，转换为Prometheus metrics格式，并暴露端口让Prometheus采集。

**说明：官方不推荐使用第一种方式，一方面配置复杂，另一方面因为它需要一个单独的进程，而这个进程本身的监控又成了新的问题，所以本文重点围绕第二种用法讲如何在K8S 环境下使用JMX Exporter暴露JVM监控指标。**

--------------------

# 三、环境信息 #

**1、监控系统架构图**  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70]  
**2、环境说明**

**业务系统涉及到的java 服务，有些是tomcat类型，有些是微服务，微服务采用的是k8s来部署的。由于公司涉及到的系统有多个，并不是所有的的业务系统都有使用k8s，所以起初为了统一管理监控平台，专门申请了一台ECS服务器来部署prometheus和grafana，也就是说所有涉及到的监控数据都通过这台ECS服务器来存储和展示，针对不同监控类型数据部署了多个prometheus。**

--------------------

# 四、部署操作 #

### 4.1、资源下载 ###

**[jmx\_prometheus\_javaagent-0.16.1.jar和prometheus-jmx-config.yaml][jmx_prometheus_javaagent-0.16.1.jar_prometheus-jmx-config.yaml]**  
![在这里插入图片描述][20210719133753612.png]

--------------------

### 4.2、修改Dockerfile ###

# 修改前
    FROM java:8u111-jdk
    RUN cp /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && \
        echo 'Asia/Shanghai' >/etc/timezone
    ADD mpmt-cp-adapter.jar adapter.jar
    EXPOSE 8083
    ENTRYPOINT ["java","-Xmx2048m","-Xms512m","-XX:MaxMetaspaceSize=512m","-XX:MetaspaceSize=256m","-Djava.security.egd=file:/dev/./urandom","-jar","/adapter.jar","--spring.profiles.active=test"]
    
    # 修改后
    FROM java:8u111-jdk
    RUN cp /usr/share/zoneinfo/Asia/Shanghai /etc/localtime && \
        echo 'Asia/Shanghai' >/etc/timezone
    ADD mpmt-cp-adapter.jar adapter.jar
    ADD prometheus-jmx-config.yaml /prometheus-jmx-config.yaml
    ADD jmx_prometheus_javaagent-0.16.1.jar /jmx_prometheus_javaagent-0.16.1.jar
    EXPOSE 8083
    EXPOSE 7070
    ENTRYPOINT ["java","-javaagent:/jmx_prometheus_javaagent-0.16.1.jar=7070:prometheus-jmx-config.yaml","-Xmx2048m","-Xms512m","-XX:MaxMetaspaceSize=512m","-XX:MetaspaceSize=256m","-Djava.security.egd=file:/dev/./urandom","-jar","/adapter.jar","--spring.profiles.active=test"]

--------------------

### 4.3、修改mpmt-adapter.yaml文件 ###

# 修改前
    ---
    apiVersion: v1
    kind: Service
    metadata:
      name: mpmt-adapter-svc
      labels:
        app: mpmt-adapter
    spec:
      type: NodePort
      ports:
      - port: 8083
        targetPort: 8083
        protocol: TCP
        nodePort: 31087
      selector:
        app: mpmt-adapter
    ---
    apiVersion: apps/v1
    kind: StatefulSet
    metadata:
      name: mpmt-adapter
    
    spec:
      serviceName: "mpmt-adapter"
      podManagementPolicy: Parallel    
      replicas: 1
      updateStrategy:
        type: RollingUpdate
      selector:
        matchLabels:
          app: mpmt-adapter
      template:
        metadata:
          labels:
            app: mpmt-adapter
        spec:
          containers:
          - name: mpmt-adapter
            image: server.harbor.com:8888/library/mpmt-adapter:202107191029
            imagePullPolicy: IfNotPresent
            ports:
            - containerPort: 8083
            resources:
              requests:
                cpu: "500m"
                memory: "4Gi"
              limits:
                cpu: "1000m"
                memory: "6Gi"
    
    # 修改后
    ---
    apiVersion: v1
    kind: Service
    metadata:
      name: mpmt-adapter-svc
      labels:
        app: mpmt-adapter
    spec:
      type: NodePort
      ports:
      - port: 8083
        targetPort: 8083
        protocol: TCP
        nodePort: 31087
        name: jdk   #增加内容 
      - port: 7070  #增加内容 
        targetPort: 7070 #增加内容 
        protocol: TCP #增加内容 
        nodePort: 35001 #增加内容 
        name: jvm #增加内容 
      selector:
        app: mpmt-adapter
    ---
    apiVersion: apps/v1
    kind: StatefulSet
    metadata:
      name: mpmt-adapter
    
    spec:
      serviceName: "mpmt-adapter"
      podManagementPolicy: Parallel
      replicas: 1
      updateStrategy:
        type: RollingUpdate
      selector:
        matchLabels:
          app: mpmt-adapter
      template:
        metadata:
          labels:
            app: mpmt-adapter
        spec:
          containers:
          - name: mpmt-adapter
            image: server.harbor.com:8888/library/mpmt-adapter:202107191029
            imagePullPolicy: IfNotPresent
            ports:
            - containerPort: 8083
            resources:
              requests:
                cpu: "500m"
                memory: "4Gi"
              limits:
                cpu: "1000m"
                memory: "6Gi"

**说明：这里由于prometheus部署在另外一台非k8s集群主机上，所以这里的微服务的jvm端口都采用主机端口映射方式，要注意的是如果同一个yaml文件中需要映射多个端口到宿主机上，需要对不通的映射端口添加名称以区分，这里采用的jdk和jvm的方式。**

--------------------

### 4.4、部署prometheus ###

mkdir -pv /data/pkgs/prometheus/jvm-exporter/etc/conf.d/rules
    vim /data/pkgs/prometheus/jvm-exporter/etc/conf.d/rules/rule_jvm_alert.yml
    groups:
      - name: jvm-alerting
        rules:
        # down了超过1分钟
        - alert: instance-down
          expr: up == 0
          for: 1m
          labels:
            severity: 严重
            team: 运维
          annotations:
            summary: "Instance { { $labels.instance }} down"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been down for more than 1 minutes."
    
        # down了超过5分钟
        - alert: instance-down
          expr: up == 0
          for: 5m
          labels:
            severity: 灾难
            team: 运维
          annotations:
            summary: "Instance { { $labels.instance }} down"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been down for more than 5 minutes."
    
        # 堆空间使用超过50%
        - alert: heap-usage-too-much
          expr: jvm_memory_bytes_used{ job="jvm-exporter", area="heap"} / jvm_memory_bytes_max * 100 > 50
          for: 1m
          labels:
            severity: 警告
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} memory usage > 50%"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [heap usage > 50%] for more than 1 minutes. current usage ({ { $value }}%)"
    
        # 堆空间使用超过80%
        - alert: heap-usage-too-much
          expr: jvm_memory_bytes_used{ job="jvm-exporter", area="heap"} / jvm_memory_bytes_max * 100 > 80
          for: 1m
          labels:
            severity: 严重
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} memory usage > 80%"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [heap usage > 80%] for more than 1 minutes. current usage ({ { $value }}%)"
        
        # 堆空间使用超过90%
        - alert: heap-usage-too-much
          expr: jvm_memory_bytes_used{ job="jvm-exporter", area="heap"} / jvm_memory_bytes_max * 100 > 90
          for: 1m
          labels:
            severity: 灾难
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} memory usage > 90%"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [heap usage > 90%] for more than 1 minutes. current usage ({ { $value }}%)"
    
        # 在5分钟里，Old GC花费时间超过30%
        - alert: old-gc-time-too-much
          expr: increase(jvm_gc_collection_seconds_sum{ gc="PS MarkSweep"}[5m]) > 5 * 60 * 0.3
          for: 5m
          labels:
            severity: 警告
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} Old GC time > 30% running time"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [Old GC time > 30% running time] for more than 5 minutes. current seconds ({ { $value }}%)"
    
        # 在5分钟里，Old GC花费时间超过50% 
        - alert: old-gc-time-too-much
          expr: increase(jvm_gc_collection_seconds_sum{ gc="PS MarkSweep"}[5m]) > 5 * 60 * 0.5
          for: 5m
          labels:
            severity: 严重
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} Old GC time > 50% running time"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [Old GC time > 50% running time] for more than 5 minutes. current seconds ({ { $value }}%)"
    
        # 在5分钟里，Old GC花费时间超过80%
        - alert: old-gc-time-too-much
          expr: increase(jvm_gc_collection_seconds_sum{ gc="PS MarkSweep"}[5m]) > 5 * 60 * 0.8
          for: 5m
          labels:
            severity: 灾难
            team: 运维
          annotations:
            summary: "JVM Instance { { $labels.instance }} Old GC time > 80% running time"
            description: "{ { $labels.instance }} of job { { $labels.job }} has been in status [Old GC time > 80% running time] for more than 5 minutes. current seconds ({ { $value }}%)"
    
    
    # 4、准备lflk-test.json
    vim /data/pkgs/yibot_lflk_tools_v4.0/prometheus/jvm-exporter/etc/conf.d/lflk-test.json
    [
      { 
        "targets": [
          "172.181.111.110:35001"
        ],
        "labels": { 
          "tag": "lflk",
          "system": "大地分局调度系统",
          "env": "测试环境",
          "owner": "大地分局",
          "cloud": "华为云",
          "ip": "10.96.30.87",
          "java": "mpmt-adapter"
        },
      { 
        "targets": [
          "172.181.111.110:35002"
        ],
        "labels": { 
          "tag": "lflk",
          "system": "大地分局调度系统",
          "env": "测试环境",
          "owner": "大地分局",
          "cloud": "华为云",
          "ip": "10.96.30.87",
          "java": "mpmt-watcher"
        }
    ]
    
    # 4、启动容器
    docker run -d --name prometheus-jvm-exporter -p 9101:9090  -v /etc/localtime:/etc/localtime \
     -v /data/pkgs/yibot_lflk_tools_v4.0/prometheus/jvm-exporter/etc/prometheus.yml:/etc/prometheus/prometheus.yml \
     -v /data/pkgs/yibot_lflk_tools_v4.0/prometheus/jvm-exporter/etc/conf.d:/etc/prometheus/conf.d \
     prom/prometheus --web.enable-lifecycle  --config.file=/etc/prometheus/prometheus.yml
    
    # 5、测试
    # 查看mpmt-adapter服务jvm数据
    http://172.181.111.110:35001
    
    # 查看Prometheus是否获取微服务jvm数据
    http://172.181.111.107:9101

**如下图所示，则表示Prometheus数据**  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 1]  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 2]

### 4.5、导入dashborard模板 ###

**说明：默认是grafana已经安装，至于如何导入模板及添加数据源就不做介绍了。**

# 官方模板
    https://grafana.com/grafana/dashboards/12856/revisions
    
    # 个人修改模板
    https://download.csdn.net/download/m0_37814112/20374260

**JVM监控效果图如下：**  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 3]![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 4]  
**钉钉告警效果图如下：**  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 5]

--------------------

# 总结：整理不易，如果对你有帮助，可否点赞关注一下？ #

更多详细内容请参考：[企业级K8s集群运维实战][K8s]

[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70]: /images/20220829/3ebba13103f04407ac504c4d5936f08e.png
[jmx_prometheus_javaagent-0.16.1.jar_prometheus-jmx-config.yaml]: https://download.csdn.net/download/m0_37814112/20364119?spm=1001.2014.3001.5501
[20210719133753612.png]: /images/20220829/9410b3d1c4324e63832ddf1f0923447e.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 1]: /images/20220829/30626999d73644b9ad37cb8f18ec8cc4.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 2]: /images/20220829/ec0b8bf187ac4ab1868cf621aa223549.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20210720140921459.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 4]: /images/20220829/0eabe16cd2654bf1a06ea9420673adcb.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L20wXzM3ODE0MTEy_size_16_color_FFFFFF_t_70 5]: /images/20220829/568ea17d197d48e68f50ce7a4f66b8b6.png
[K8s]: https://blog.csdn.net/m0_37814112/category_10856504.html?spm=1001.2014.3001.5482