當(dāng)我們有多個Kubernetes集群的時候，這個時候就需要需要指標(biāo)匯總的需求了，如上圖一樣，我們假定在外部部署一個Prometheus的Federate，然后去采集當(dāng)前k8s中的kube-system與default倆個 namespace。

環(huán)境

我的本地環(huán)境使用的 sealos 一鍵部署，主要是為了便于測試。

部署 Prometheus聯(lián)邦集群

創(chuàng)建prometheus-federate數(shù)據(jù)目錄

# 在m1上執(zhí)行
mkdir /data/prometheus-federate/
chown -R 65534:65534 /data/prometheus-federate/

創(chuàng)建Prometheus聯(lián)邦 StorageClass 配置文件

cd /data/manual-deploy/prometheus/
cat prometheus-federate-storageclass.yaml
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: prometheus-federate-lpv
provisioner: kubernetes.io/no-provisioner
volumeBindingMode: WaitForFirstConsumer

創(chuàng)建Prometheus聯(lián)邦pv配置文件

apiVersion: v1
kind: PersistentVolume
metadata:
  name: prometheus-federate-lpv-0
spec:
  capacity:
    storage: 10Gi
  volumeMode: Filesystem
  accessModes:
  - ReadWriteOnce
  persistentVolumeReclaimPolicy: Retain
  storageClassName: prometheus-federate-lpv
  local:
    path: /data/prometheus-federate
  nodeAffinity:
    required:
      nodeSelectorTerms:
      - matchExpressions:
        - key: kubernetes.io/hostname
          operator: In
          values:
          - sealos-k8s-m1

創(chuàng)建Prometheus聯(lián)邦configmap配置文件

cat prometheus-federate-configmap.yaml
apiVersion: v1
kind: ConfigMap
metadata:
  name: prometheus-federate-config
  namespace: kube-system
data:
  alertmanager_rules.yaml: |
    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: (node_memory_MemTotal_bytes -(node_memory_MemFree_bytes+node_memory_Buffers_bytes+node_memory_Cached_bytes )) / node_memory_MemTotal_bytes * 100 > 80
        for: 1m
        labels:
          team: ops
        annotations:
          summary: "cluster:{{ $labels.cluster }} {{ $labels.instance }}: High Memory usage detected"
          description: "{{ $labels.instance }}: Memory usage is above 55% (current value is: {{ $value }}"
  prometheus.yml: |
    global:
      scrape_interval:     30s
      evaluation_interval: 30s
    alerting:
      alertmanagers:
      - static_configs:
        - targets:
            - alertmanager-0.alertmanager-operated:9093
            - alertmanager-1.alertmanager-operated:9093      
    rule_files:
      - "/etc/prometheus/alertmanager_rules.yaml"
    scrape_configs:
      - job_name: 'federate'
        scrape_interval: 30s
        honor_labels: true
        metrics_path: '/federate'
        params:
          'match[]':
            - '{job=~"kubernetes.*"}'
            - '{job="prometheus"}'
        static_configs:
          - targets:
            - 'prometheus-0.prometheus:9090'
            - 'prometheus-1.prometheus:9090'
            - 'prometheus-2.prometheus:9090'

創(chuàng)建Prometheus聯(lián)邦的statefulse文件

cat prometheus-federate-statefulset.yaml
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: prometheus-federate
  namespace: kube-system
  labels:
    k8s-app: prometheus-federate
    kubernetes.io/cluster-service: "true"
spec:
  serviceName: "prometheus-federate"
  podManagementPolicy: "Parallel"
  replicas: 1
  selector:
    matchLabels:
      k8s-app: prometheus-federate
  template:
    metadata:
      labels:
        k8s-app: prometheus-federate
      annotations:
        scheduler.alpha.kubernetes.io/critical-pod: ''
    spec:
      affinity:
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:
          - labelSelector:
              matchExpressions:
              - key: k8s-app
                operator: In
                values:
                - prometheus-federate
            topologyKey: "kubernetes.io/hostname"
      priorityClassName: system-cluster-critical
      hostNetwork: true
      DNSPolicy: ClusterFirstWithHostNet
      containers:
      - name: prometheus-federate-configmap-reload
        image: "jimmidyson/configmap-reload:v0.4.0"
        imagePullPolicy: "IfNotPresent"
        args:
          - --volume-dir=/etc/config
          - --webhook-url=http://localhost:9091/-/reload
        volumeMounts:
          - name: config-volume
            mountPath: /etc/config
            readOnly: true
        resources:
          limits:
            cpu: 10m
            memory: 10Mi
          requests:
            cpu: 10m
            memory: 10Mi
        securityContext:
            runAsUser: 0
            privileged: true
      - image: prom/prometheus:v2.20.0
        imagePullPolicy: IfNotPresent
        name: prometheus
        command:
          - "/bin/prometheus"
        args:
          - "--web.listen-address=0.0.0.0:9091"
          - "--config.file=/etc/prometheus/prometheus.yml"
          - "--storage.tsdb.path=/prometheus"
          - "--storage.tsdb.retention=24h"
          - "--web.console.libraries=/etc/prometheus/console_libraries"
          - "--web.console.templates=/etc/prometheus/consoles"
          - "--web.enable-lifecycle"
        ports:
          - containerPort: 9091
            protocol: TCP
        volumeMounts:
          - mountPath: "/prometheus"
            name: prometheus-federate-data
          - mountPath: "/etc/prometheus"
            name: config-volume
        readinessProbe:
          httpGet:
            path: /-/ready
            port: 9091
          initialDelaySeconds: 30
          timeoutSeconds: 30
        livenessProbe:
          httpGet:
            path: /-/healthy
            port: 9091
          initialDelaySeconds: 30
          timeoutSeconds: 30
        resources:
          requests:
            cpu: 100m
            memory: 100Mi
          limits:
            cpu: 1000m
            memory: 2500Mi
        securityContext:
            runAsUser: 0
            privileged: true
      serviceAccountName: prometheus
      volumes:
        - name: config-volume
          configMap:
            name: prometheus-federate-config
  volumeClaimTemplates:
    - metadata:
        name: prometheus-federate-data
      spec:
        accessModes: [ "ReadWriteOnce" ]
        storageClassName: "prometheus-federate-lpv"
        resources:
          requests:
            storage: 5Gi       

創(chuàng)建Prometheus聯(lián)邦的svc文件

cat prometheus-service-statefulset.yaml
apiVersion: v1
kind: Service
metadata:
  name: prometheus
  namespace: kube-system
spec:
  ports:
    - name: prometheus
      port: 9090
      targetPort: 9090
  selector:
    k8s-app: prometheus
  clusterIP: None

部署

cd /data/manual-deploy/prometheus/
prometheus-federate-configmap.yaml
prometheus-federate-pv.yaml
prometheus-federate-service-statefulset.yaml
prometheus-federate-statefulset.yaml
prometheus-federate-storageclass.yaml
kubectl apply -f prometheus-federate-storageclass.yaml
kubectl apply -f prometheus-federate-pv.yaml
kubectl apply -f prometheus-federate-configmap.yaml
kubectl apply -f prometheus-federate-statefulset.yaml
kubectl apply -f prometheus-federate-service-statefulset.yaml

驗證

# pv
kubectl -n kube-system get pvc |grep federate
prometheus-federate-data-prometheus-federate-0   Bound    prometheus-federate-lpv-0   10Gi RWO   prometheus-federate-lpv 4h
kubectl -n kube-system get pod |grep federate
prometheus-federate-0                      2/2     Running   0          2d4h

對此，聯(lián)邦的配置就完成了，可以在瀏覽器中訪問192.168.1.151:9091查看相應(yīng)的targets信息，以及配置的rules規(guī)則，觸發(fā)下警報，看看Alertmanager集群已經(jīng)部署成功了。