随笔- 542 文章- 6 评论- 27 阅读- 29万

Kubernetes——StatefulSet资源升级（滚动更新操作、暂存更新操作）

StatefulSet资源升级

自 Kubernetes 1.7 版本起，StatefulSet 资源支持自动更新机制，其更新策略由 spec.updateStrategy 字段定义，默认为 RollingUpdate，即滚动更新。

一、滚动更新操作

滚动更新 StatefulSet 控制器的 Pod 资源以逆序的形式从其最大索引编号的 Pod 资源逐一进行，它在终止一个 Pod 资源、更新资源并待其就绪后启动更新下一个资源，即索引号比当前号小 1 的 Pod 资源。对于主从复制类的集群应用来说，这样也能保证起主节点作用的 Pod 资源最后进行更新，确保兼容性。

StatefulSet 的默认更新策略为滚动更新，通过 "kubectl get statefulset NAME -o yaml" 命令中的输出可以获取相关的信息：

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

[root@k8s-master01-test-2-26 ~]# kubectl get statefulset alertmanager-main -n kubesphere-monitoring-system
NAME                READY   AGE
alertmanager-main   1/1     66m
[root@k8s-master01-test-2-26 ~]# kubectl get statefulset alertmanager-main -n kubesphere-monitoring-system -o yaml
apiVersion: apps/v1
kind: StatefulSet
metadata:
  annotations:
    prometheus-operator-input-hash: "10566271792385409484"
  creationTimestamp: "2022-06-29T05:31:54Z"
  generation: 1
  labels:
    app.kubernetes.io/component: alert-router
    app.kubernetes.io/instance: main
    app.kubernetes.io/name: alertmanager
    app.kubernetes.io/part-of: kube-prometheus
    app.kubernetes.io/version: 0.23.0
  name: alertmanager-main
  namespace: kubesphere-monitoring-system
  ownerReferences:
  - apiVersion: monitoring.coreos.com/v1
    blockOwnerDeletion: true
    controller: true
    kind: Alertmanager
    name: main
    uid: c054f193-4d87-49b8-b8e7-9275c753460c
  resourceVersion: "6347"
  uid: dbd3d926-fefd-4b11-94a6-079b08b7a293
spec:
  podManagementPolicy: Parallel
  replicas: 1
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      alertmanager: main
      app.kubernetes.io/instance: main
      app.kubernetes.io/managed-by: prometheus-operator
      app.kubernetes.io/name: alertmanager
  serviceName: alertmanager-operated
  template:
    metadata:
      annotations:
        kubectl.kubernetes.io/default-container: alertmanager
      creationTimestamp: null
      labels:
        alertmanager: main
        app.kubernetes.io/component: alert-router
        app.kubernetes.io/instance: main
        app.kubernetes.io/managed-by: prometheus-operator
        app.kubernetes.io/name: alertmanager
        app.kubernetes.io/part-of: kube-prometheus
        app.kubernetes.io/version: 0.23.0
    spec:
      affinity:
        podAntiAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:
          - podAffinityTerm:
              labelSelector:
                matchExpressions:
                - key: alertmanager
                  operator: In
                  values:
                  - main
              namespaces:
              - kubesphere-monitoring-system
              topologyKey: kubernetes.io/hostname
            weight: 100
      containers:
      - args:
        - --config.file=/etc/alertmanager/config/alertmanager.yaml
        - --storage.path=/alertmanager
        - --data.retention=120h
        - --cluster.listen-address=
        - --web.listen-address=:9093
        - --web.route-prefix=/
        - --cluster.peer=alertmanager-main-0.alertmanager-operated:9094
        - --cluster.reconnect-timeout=5m
        env:
        - name: POD_IP
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: status.podIP
        image: registry.cn-beijing.aliyuncs.com/kubesphereio/alertmanager:v0.23.0
        imagePullPolicy: IfNotPresent
        livenessProbe:
          failureThreshold: 10
          httpGet:
            path: /-/healthy
            port: web
            scheme: HTTP
          periodSeconds: 10
          successThreshold: 1
          timeoutSeconds: 3
        name: alertmanager
        ports:
        - containerPort: 9093
          name: web
          protocol: TCP
        - containerPort: 9094
          name: mesh-tcp
          protocol: TCP
        - containerPort: 9094
          name: mesh-udp
          protocol: UDP
        readinessProbe:
          failureThreshold: 10
          httpGet:
            path: /-/ready
            port: web
            scheme: HTTP
          initialDelaySeconds: 3
          periodSeconds: 5
          successThreshold: 1
          timeoutSeconds: 3
        resources:
          limits:
            cpu: 200m
            memory: 200Mi
          requests:
            cpu: 20m
            memory: 30Mi
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            drop:
            - ALL
          readOnlyRootFilesystem: true
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: FallbackToLogsOnError
        volumeMounts:
        - mountPath: /etc/alertmanager/config
          name: config-volume
        - mountPath: /etc/alertmanager/certs
          name: tls-assets
          readOnly: true
        - mountPath: /alertmanager
          name: alertmanager-main-db
      - args:
        - --listen-address=:8080
        - --reload-url=http://localhost:9093/-/reload
        - --watched-dir=/etc/alertmanager/config
        command:
        - /bin/prometheus-config-reloader
        env:
        - name: POD_NAME
          valueFrom:
            fieldRef:
              apiVersion: v1
              fieldPath: metadata.name
        - name: SHARD
          value: "-1"
        image: registry.cn-beijing.aliyuncs.com/kubesphereio/prometheus-config-reloader:v0.55.1
        imagePullPolicy: IfNotPresent
        name: config-reloader
        ports:
        - containerPort: 8080
          name: reloader-web
          protocol: TCP
        resources:
          limits:
            cpu: 100m
            memory: 50Mi
          requests:
            cpu: 100m
            memory: 50Mi
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            drop:
            - ALL
          readOnlyRootFilesystem: true
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: FallbackToLogsOnError
        volumeMounts:
        - mountPath: /etc/alertmanager/config
          name: config-volume
          readOnly: true
      dnsPolicy: ClusterFirst
      nodeSelector:
        kubernetes.io/os: linux
      restartPolicy: Always
      schedulerName: default-scheduler
      securityContext:
        fsGroup: 2000
        runAsNonRoot: true
        runAsUser: 1000
      serviceAccount: alertmanager-main
      serviceAccountName: alertmanager-main
      terminationGracePeriodSeconds: 120
      volumes:
      - name: config-volume
        secret:
          defaultMode: 420
          secretName: alertmanager-main-generated
      - name: tls-assets
        projected:
          defaultMode: 420
          sources:
          - secret:
              name: alertmanager-main-tls-assets-0
      - emptyDir: {}
        name: alertmanager-main-db
  updateStrategy:
    type: RollingUpdate
status:
  availableReplicas: 1
  collisionCount: 0
  currentReplicas: 1
  currentRevision: alertmanager-main-cd5bc8fdc
  observedGeneration: 1
  readyReplicas: 1
  replicas: 1
  updateRevision: alertmanager-main-cd5bc8fdc
  updatedReplicas: 1
[root@k8s-master01-test-2-26 ~]# 

"kubectl rollout status" 命令跟踪 StatefulSet 资源滚动更新过程中的状态信息。

二、暂存更新操作

当用户需要设定一个更新操作，但又不希望它立即执行时，可将更新操作予以 "暂存"，待条件满足后再手动触发其执行更新。

StatefulSet 资源的分区更新机制能够实现此项功能。在设定更新操作之前，将 .spec.updateStrategy.rollingUpdate.partition 字段的值设置为 Pod 资源的副本数量，即比 Pod 资源的最大索引号大 1，这就意味着，所有的 Pod 资源都不会处于可直接更新的分区之内，那么于其后设定的更新操作也就不会真正执行，直到用户降低分区编号至现有 Pod 资源索引号范围之内。

[root@k8s-master01-test-2-26 ~]# kubectl explain statefulset.spec.updateStrategy.rollingUpdate.partition
KIND:     StatefulSet
VERSION:  apps/v1
 
FIELD:    partition <integer>
 
DESCRIPTION:
     Partition indicates the ordinal at which the StatefulSet should be
     partitioned for updates. During a rolling update, all pods from ordinal
     Replicas-1 to Partition are updated. All pods from ordinal Partition-1 to 0
     remain untouched. This is helpful in being able to do a canary based
     deployment. The default value is 0.
[root@k8s-master01-test-2-26 ~]# 