Prometheus監(jiān)控k8s

[toc]

k8s監(jiān)控方案

cadvisor+heapster+influxdb+grafana

缺點：只能支持監(jiān)控容器資源，無法支持業(yè)務(wù)監(jiān)控，擴展性較差

cadvisor/exporter+prometheus+grafana

總體流程: 數(shù)據(jù)采集-->匯總-->處理-->存儲-->展示

  • 容器的監(jiān)控
    • prometheus使用cadvisor采集容器監(jiān)控指標(biāo)，cadvisor集成在k8s的kubelet中-通過prometheus進程存儲-使用grafana進行展現(xiàn)
    • node的監(jiān)控-通過node_pxporter采集當(dāng)前主機的資源-通過prometheus進程存儲-使用grafana進行展現(xiàn)
    • master的監(jiān)控-通過kube-state-metrics插件從k8s中獲取到apiserver的相關(guān)數(shù)據(jù)-通過prometheus進程存儲-使用grafana進行展現(xiàn)

kubernetes監(jiān)控指標(biāo)

kubernetes自身的監(jiān)控

  1. node的資源利用率-node節(jié)點上的cpu、內(nèi)存、硬盤、鏈接
  2. node的數(shù)量-node數(shù)量與資源利用率、業(yè)務(wù)負載的比例情況、成本、資源擴展的評估
  3. pod的數(shù)量-當(dāng)負載到一定程度時，node與pod的數(shù)量，評估負載到哪個階段，大約需要多少服務(wù)器，每個pod的資源占用率如何，進行整體評估
  4. 資源對象狀態(tài)-k8s在運行過程中，會創(chuàng)建很多pod，控制器，任務(wù)，這些內(nèi)容都是由k8s中的資源對象進行維護，需要進行對資源對象的監(jiān)控，獲取資源對象的狀態(tài)

pod監(jiān)控

  1. 每個項目中pod的數(shù)量-正常的pod數(shù)量，有問題的pod數(shù)量
  2. 容器資源利用率-統(tǒng)計當(dāng)前pod的資源利用率，統(tǒng)計pod中的容器資源利用率，cpu、網(wǎng)絡(luò)、內(nèi)存評估
  3. 應(yīng)用程序-項目中的程序的自身情況，如并發(fā)，請求響應(yīng)，項目用戶數(shù)量，訂單數(shù)等

實現(xiàn)思路

監(jiān)控指標(biāo)        具體實現(xiàn)            舉例
pod性能           cadvisor            容器的cpu、內(nèi)存利用率
node性能      node-exporter       node節(jié)點的cpu、內(nèi)存利用率
k8s資源對象     kube-state-metrics  pod/deployment/service

服務(wù)發(fā)現(xiàn)
從kubernetes的api中去發(fā)現(xiàn)抓取的目標(biāo)，并始終與kubernetes集群狀態(tài)保持一致，
動態(tài)的獲取被抓取的目標(biāo)，實時的從api中獲取當(dāng)前狀態(tài)是否存在，

官方文檔
https://prometheus.io/docs/prometheus/latest/configuration/configuration/#kubernetes_sd_config

自動發(fā)現(xiàn)支持的組件：

  • node-自動發(fā)現(xiàn)集群中的node節(jié)點
  • pod-自動發(fā)現(xiàn)運行的容器和端口
  • service-自動發(fā)現(xiàn)創(chuàng)建的serviceIP、端口
  • endpoints-自動發(fā)現(xiàn)pod中的容器
  • ingress-自動發(fā)現(xiàn)創(chuàng)建的訪問入口和規(guī)則

使用prometheus監(jiān)控k8s

在k8s中部署prometheus

官方部署文檔: https://github.com/kubernetes/kubernetes/tree/master/cluster/addons/prometheus

制作prometheus PV/PVC

#安裝依賴包
yum -y install nfs-utils rpcbind

#開機啟動,
systemctl enable rpcbind.service 
systemctl enable nfs-server.service
systemctl start rpcbind.service #端口是111
systemctl start nfs-server.service # 端口是 2049 

# 創(chuàng)建一個/data/pvdata的共享目錄
# mkdir /data/pvdata
# chown nfsnobody:nfsnobody /data/pvdata
# cat /etc/exports
/data/pvdata 172.22.22.0/24(rw,async,all_squash)
# exportfs -rv
exporting 172.22.22.0/24:/data/pvdata

下載prometheus yaml部署文件

mkdir /data/k8s/yaml/kube-system/prometheus
cd /data/k8s/yaml/kube-system/prometheus/

# 從github官網(wǎng)下載yaml部署文件
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/prometheus-rbac.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/prometheus-configmap.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/prometheus-service.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/prometheus-statefulset.yaml

修改statefulset.yaml

# 刪掉最下面的10行
  volumeClaimTemplates:
  - metadata:
      name: prometheus-data
    spec:
      storageClassName: standard
      accessModes:
        - ReadWriteOnce
      resources:
        requests:
          storage: "16Gi"

# 新增下面3行
        - name: prometheus-data
          persistentVolumeClaim:  
            claimName: prometheus-data

新增pv/pvc yaml文件

mkdir /data/pvdata/prometheus
chown nfsnobody. /data/pvdata/prometheus

cat > prometheus-pvc-data.yaml << EFO
apiVersion: v1
kind: PersistentVolume
metadata:
  name: prometheus-data
spec:
  storageClassName: prometheus-data
  capacity: 
    storage: 10Gi  
  accessModes: 
    - ReadWriteOnce  
  persistentVolumeReclaimPolicy: Recycle 
  nfs:
    path: /data/pvdata/prometheus
    server: 192.168.1.155

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: prometheus-data
  namespace: kube-system  
spec:
  accessModes:
    - ReadWriteOnce 
  resources:
    requests:
      storage: 10Gi 
  storageClassName: prometheus-data
EFO

新增Prometheus-ingress.yaml文件

主要是方便外部grafana使用

cat > prometheus-ingress.yaml << EFO
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: prometheus-ingress
  namespace: kube-system
spec:
  rules:
  - host: prometheus.baiyongjie.com
    http:
      paths:
      - backend:
          serviceName: prometheus
          servicePort: 9090
EFO

應(yīng)用yaml文件

# 部署順序
1. prometheus-rbac.yaml-對prometheus訪問kube-apiserver進行授權(quán)
2. prometheus-configmap.yaml-管理prometheus主配置文件
3. prometheus-service.yaml-將prometheus暴露出去，可以訪問
4  prometheus-ingress.yaml-對外提供服務(wù)
4. prometheus-pvc-data.yaml-為pod提供數(shù)據(jù)存儲
5. prometheus-statefulset.yaml-通過有狀態(tài)的形式，將prometheus去部署
6. prometheus-ingress.yaml-對外提供服務(wù)

# 應(yīng)用yaml文件
kubectl apply -f prometheus-rbac.yaml 
kubectl apply -f prometheus-configmap.yaml 
kubectl apply  -f prometheus-ingress.yaml
kubectl apply -f prometheus-pvc-data.yaml
kubectl apply -f prometheus-service.yaml
kubectl apply -f prometheus-statefulset.yaml

# 查看部署情況
[root@master prometheus]# kubectl get pv
NAME              CAPACITY   ACCESS MODES   RECLAIM POLICY   STATUS   CLAIM                         STORAGECLASS      REASON   AGE
prometheus-data   10Gi       RWO            Recycle          Bound    kube-system/prometheus-data   prometheus-data            32m

[root@master prometheus]# kubectl get pvc -n kube-system 
NAME              STATUS   VOLUME            CAPACITY   ACCESS MODES   STORAGECLASS      AGE
prometheus-data   Bound    prometheus-data   10Gi       RWO            prometheus-data   33m

[root@master prometheus]# kubectl get service -n kube-system 
NAME         TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)          AGE
kube-dns     ClusterIP   10.96.0.10      <none>        53/UDP,53/TCP    12d
prometheus   NodePort    10.107.69.131   <none>        9090/TCP   57m

[root@master prometheus]# kubectl get statefulsets.apps  -n kube-system 
NAME         READY   AGE
prometheus   1/1     15m

[root@master prometheus]# kubectl get ingresses.extensions -n kube-system 
NAME                 HOSTS                       ADDRESS   PORTS   AGE
prometheus-ingress   prometheus.baiyongjie.com             80      7m3s

[root@master prometheus]# kubectl get pods -n kube-system  -o wide |grep prometheus
NAME                             READY   STATUS    RESTARTS   AGE   IP              NODE     NOMINATED NODE   READINESS GATES
prometheus-0                     2/2     Running   0          42s   10.244.1.6      node01   <none>           <none>

訪問ingress

# 修改hosts文件,添加ingress域名解析
192.168.1.156 prometheus.baiyongjie.com

然后訪問 http://prometheus.baiyongjie.com/graph
image.png

部署node-exporter

下載yaml文件

curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/node-exporter-ds.yml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/node-exporter-service.yaml

由于我們要獲取到的數(shù)據(jù)是主機的監(jiān)控指標(biāo)數(shù)據(jù)，而我們的 node-exporter 是運行在容器中的，所以我們在 Pod 中需要配置一些 Pod 的安全策略，這里我們就添加了hostPID: true、hostIPC: true、hostNetwork: true3個策略，用來使用主機的 PID namespace、IPC namespace 以及主機網(wǎng)絡(luò)，這些 namespace 就是用于容器隔離的關(guān)鍵技術(shù)，要注意這里的 namespace 和集群中的 namespace 是兩個完全不相同的概念。

另外我們還將主機的/dev、/proc、/sys這些目錄掛載到容器中，這些因為我們采集的很多節(jié)點數(shù)據(jù)都是通過這些文件夾下面的文件來獲取到的，比如我們在使用top命令可以查看當(dāng)前cpu使用情況，數(shù)據(jù)就來源于文件/proc/stat，使用free命令可以查看當(dāng)前內(nèi)存使用情況，其數(shù)據(jù)來源是來自/proc/meminfo文件。

另外由于我們集群使用的是 kubeadm 搭建的，所以如果希望 master 節(jié)點也一起被監(jiān)控，則需要添加響應(yīng)的容忍。

// 修改node-exporter-ds.yml文件
添加
    spec:
      hostPID: true 
      hostIPC: true
      hostNetwork: true
      
      tolerations:
      - key: "node-role.kubernetes.io/master"
        operator: "Exists"
        effect: "NoSchedule"
        
      volumes:
        - name: proc
          hostPath:
            path: /proc
        - name: dev
          hostPath:
            path: /dev
        - name: sys
          hostPath:
            path: /sys
        - name: rootfs
          hostPath:
            path: /

應(yīng)用yaml文件

kubectl apply -f node-exporter-service.yaml
kubectl apply -f node-exporter-ds.yml 

# 查看部署情況
[root@master prometheus]# kubectl get pods -n kube-system |grep node-export 
node-exporter-lb7gb              1/1     Running   0          4m59s
node-exporter-q22zn              1/1     Running   0          4m59s

[root@master prometheus]# kubectl get service -n kube-system |grep node-export      
node-exporter   ClusterIP   None            <none>        9100/TCP        5m49s

查看Prometheus是否獲取到數(shù)據(jù)

image.png

部署kube-state-metrics

下載yaml文件

curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/kube-state-metrics-service.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/kube-state-metrics-rbac.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/kube-state-metrics-deployment.yaml

應(yīng)用yaml文件

kubectl apply -f kube-state-metrics-service.yaml
kubectl apply -f kube-state-metrics-rbac.yaml
kubectl apply -f kube-state-metrics-deployment.yaml

部署grafana

生成yaml文件

grafana-pvc.yaml

mkdir /data/pvdata/prometheus-grafana
chown nfsnobody. /data/pvdata/prometheus-grafana

cat > grafana-pvc.yaml << EFO
apiVersion: v1
kind: PersistentVolume
metadata:
  name: prometheus-grafana
spec:
  storageClassName: prometheus-grafana
  capacity: 
    storage: 2Gi  
  accessModes: 
    - ReadWriteOnce  
  persistentVolumeReclaimPolicy: Recycle 
  nfs:
    path: /data/pvdata/prometheus-grafana
    server: 192.168.1.155

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: prometheus-grafana
  namespace: kube-system  
spec:
  accessModes:
    - ReadWriteOnce 
  resources:
    requests:
      storage: 2Gi 
  storageClassName: prometheus-grafana
EFO

grafana-ingress.yaml

cat > grafana-ingress.yaml << EFO
apiVersion: extensions/v1beta1
kind: Ingress
metadata:
   name: grafana
   namespace: kube-system
spec:
   rules:
   - host: grafana.baiyongjie.com
     http:
       paths:
       - path: /
         backend:
          serviceName: grafana
          servicePort: 3000
EFO

grafana-deployment.yaml

# cat > grafana-deployment.yaml << EFO
apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: grafana
  namespace: kube-system
  labels:
    app: grafana
spec:
  revisionHistoryLimit: 10
  template:
    metadata:
      labels:
        app: grafana
        component: prometheus
    spec:
      containers:
      - name: grafana
        env:
        - name: GF_SECURITY_ADMIN_USER
          value: admin
        - name: GF_SECURITY_ADMIN_PASSWORD
          value: admin
        image: grafana/grafana:5.3.0
        imagePullPolicy: IfNotPresent
        ports:
        - containerPort: 3000
          name: grafana
        readinessProbe:
          failureThreshold: 10
          httpGet:
            path: /api/health
            port: 3000
            scheme: HTTP
          initialDelaySeconds: 30
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 30
        livenessProbe:
          failureThreshold: 3
          httpGet:
            path: /api/health
            port: 3000
            scheme: HTTP
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 1
        resources:
          limits:
            cpu: 100m
            memory: 256Mi
          requests:
            cpu: 100m
            memory: 256Mi
        volumeMounts:
        - mountPath: /var/lib/grafana
          subPath: grafana
          name: grafana-volumes
      volumes:
      - name: grafana-volumes
        persistentVolumeClaim:
          claimName: prometheus-grafana
EFO

部署yaml文件

kubectl apply -f grafana-pvc.yaml
kubectl apply -f grafana-ingress.yaml
kubectl apply -f grafana-deployment.yaml

# 查看部署情況
[root@master prometheus]# kubectl get service -n kube-system |grep grafana
grafana              ClusterIP   10.105.159.132   <none>        3000/TCP            150m

[root@master prometheus]# kubectl get ingresses.extensions -n kube-system |grep grafana       
grafana              grafana.baiyongjie.com                80      150m

[root@master prometheus]# kubectl get pods -n kube-system |grep grafana                     
grafana-6f6d77d98d-wwmbd              1/1     Running   0          53m

配置grafana

修改本地hosts文件添加ingress域名解析,然后訪問 http://grafana.baiyongjie.com

image.png
  • 導(dǎo)入dashboard,推薦
    • 3131 Kubernetes All Nodes
    • 3146 Kubernetes Pods
    • 8685 K8s Cluster Summary
    • 10000 Cluster Monitoring for Kubernetes
image.png
image.png
image.png

部署alertmanager

下載yaml文件

curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/alertmanager-pvc.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/alertmanager-service.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/alertmanager-deployment.yaml
curl -O https://raw.githubusercontent.com/kubernetes/kubernetes/master/cluster/addons/prometheus/alertmanager-configmap.yaml

修改yaml文件

alertmanager-pvc.yaml

mkdir /data/pvdata/prometheus-alertmanager
chown nfsnobody. /data/pvdata/prometheus-alertmanager

cat > alertmanager-pvc.yaml  << EFO
apiVersion: v1
kind: PersistentVolume
metadata:
  name: prometheus-alertmanager
spec:
  storageClassName: prometheus-alertmanager
  capacity: 
    storage: 2Gi  
  accessModes: 
    - ReadWriteOnce  
  persistentVolumeReclaimPolicy: Recycle 
  nfs:
    path: /data/pvdata/prometheus-alertmanager
    server: 192.168.1.155

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: prometheus-alertmanager
  namespace: kube-system  
spec:
  accessModes:
    - ReadWriteOnce 
  resources:
    requests:
      storage: 2Gi 
  storageClassName: prometheus-alertmanager
EFO

alertmanager-deployment.yaml

# 修改最后一行的claimName
        - name: storage-volume
          persistentVolumeClaim:
            claimName: prometheus-alertmanager

應(yīng)用yaml文件

kubectl apply -f alertmanager-pvc.yaml
kubectl apply -f alertmanager-configmap.yaml
kubectl apply -f alertmanager-service.yaml
kubectl apply -f alertmanager-deployment.yaml

# 查看部署情況
[root@master prometheus-ink8s]# kubectl get all -n kube-system  |grep alertmanager
pod/alertmanager-c564cb9fc-bfrvb          2/2     Running   0          71s
service/alertmanager         ClusterIP   10.102.208.66   <none>        80/TCP              5m44s
deployment.apps/alertmanager         1/1     1            1           71s
replicaset.apps/alertmanager-c564cb9fc          1         1         1       71s

創(chuàng)建告警規(guī)則

// 修改prometheus-configmap.yaml文件
kubectl edit configmaps prometheus-config -n kube-system 

// 在prometheus.yml: |下面添加
    alerting:
      alertmanagers:
      - static_configs:
        - targets:
          - alertmanager:80
    rule_files:
    - "/etc/config/rules.yml"


// 創(chuàng)建告警規(guī)則, 在最下面添加
  rules.yml: |
    groups:
    - name: example
      rules:
      - alert: InstanceDown
        expr: up == 0
        for: 1m
        labels:
          severity: page
        annotations:
          summary: "Instance {{ $labels.instance }} down"
          description: "{{ $labels.instance }} of job {{ $labels.job }} has been down for more than 1 minutes."
      - alert: NodeMemoryUsage
        expr: (sum(node_memory_MemTotal) - sum(node_memory_MemFree+node_memory_Buffers+node_memory_Cached) ) / sum(node_memory_MemTotal) * 100 > 20
        for: 2m
        labels:
          team: node
        annotations:
          summary: "{{$labels.instance}}: High Memory usage detected"
          description: "{{$labels.instance}}: Memory usage is above 20% (current value is: {{ $value }}"
   
//  重載配置文件


# kubectl apply -f prometheus-configmap.yaml
# kubectl get service -n kube-system |grep prometheus
prometheus           ClusterIP   10.111.97.89    <none>        9090/TCP            4h42m
# curl -X POST http://10.111.97.89:9090/-/reload

創(chuàng)建郵件告警

# 修改alertmanager-configmap.yaml文件
cat > alertmanager-configmap.yaml  << EFO
apiVersion: v1
kind: ConfigMap
metadata:
  name: alertmanager-config
  namespace: kube-system
  labels:
    kubernetes.io/cluster-service: "true"
    addonmanager.kubernetes.io/mode: EnsureExists
data:
  alertmanager.yml: |
    global:
      resolve_timeout: 3m #解析的超時時間
      smtp_smarthost: 'smtp.163.com:25'
      smtp_from: 'USERNAMR@163.com'
      smtp_auth_username: 'USERNAMR@163.com'
      smtp_auth_password: 'PASSWORD'
      smtp_require_tls: false
    
    route:
      group_by: ['example']
      group_wait: 60s
      group_interval: 60s
      repeat_interval: 12h
      receiver: 'mail'
    
    receivers:
    - name: 'mail'
      email_configs:
      - to: 'misterbyj@163.com'
        send_resolved: true
EFO

kubectl delete configmaps -n kube-system alertmanager-config 
kubectl apply  -f alertmanager-configmap.yaml

查看告警

** 訪問Prometheus, 查看是否有alerts告警規(guī)則 **


image.png
image.png
image.png
最后編輯于 ：
?著作權(quán)歸作者所有,轉(zhuǎn)載或內(nèi)容合作請聯(lián)系作者
【社區(qū)內(nèi)容提示】社區(qū)部分內(nèi)容疑似由AI輔助生成，瀏覽時請結(jié)合常識與多方信息審慎甄別。
平臺聲明：文章內(nèi)容（如有圖片或視頻亦包括在內(nèi)）由作者上傳并發(fā)布，文章內(nèi)容僅代表作者本人觀點，簡書系信息發(fā)布平臺，僅提供信息存儲服務(wù)。

相關(guān)閱讀更多精彩內(nèi)容

友情鏈接更多精彩內(nèi)容