MetalLB之从入门到放弃

简介

MetalLB是使用标准路由协议实现部署在裸机的Kubernetes集群的负载均衡器。

部署MetalLB

直接使用官网的YAML文件

apiVersion: v1
kind: Namespace
metadata:
  name: metallb-system
  labels:
    app: metallb
---
apiVersion: policy/v1beta1
kind: PodSecurityPolicy
metadata:
  labels:
    app: metallb
  name: controller
  namespace: metallb-system
spec:
  allowPrivilegeEscalation: false
  allowedCapabilities: []
  allowedHostPaths: []
  defaultAddCapabilities: []
  defaultAllowPrivilegeEscalation: false
  fsGroup:
    ranges:
    - max: 65535
      min: 1
    rule: MustRunAs
  hostIPC: false
  hostNetwork: false
  hostPID: false
  privileged: false
  readOnlyRootFilesystem: true
  requiredDropCapabilities:
  - ALL
  runAsUser:
    ranges:
    - max: 65535
      min: 1
    rule: MustRunAs
  seLinux:
    rule: RunAsAny
  supplementalGroups:
    ranges:
    - max: 65535
      min: 1
    rule: MustRunAs
  volumes:
  - configMap
  - secret
  - emptyDir
---
apiVersion: policy/v1beta1
kind: PodSecurityPolicy
metadata:
  labels:
    app: metallb
  name: speaker
  namespace: metallb-system
spec:
  allowPrivilegeEscalation: false
  allowedCapabilities:
  - NET_ADMIN
  - NET_RAW
  - SYS_ADMIN
  allowedHostPaths: []
  defaultAddCapabilities: []
  defaultAllowPrivilegeEscalation: false
  fsGroup:
    rule: RunAsAny
  hostIPC: false
  hostNetwork: true
  hostPID: false
  hostPorts:
  - max: 7472
    min: 7472
  privileged: true
  readOnlyRootFilesystem: true
  requiredDropCapabilities:
  - ALL
  runAsUser:
    rule: RunAsAny
  seLinux:
    rule: RunAsAny
  supplementalGroups:
    rule: RunAsAny
  volumes:
  - configMap
  - secret
  - emptyDir
---
apiVersion: v1
kind: ServiceAccount
metadata:
  labels:
    app: metallb
  name: controller
  namespace: metallb-system
---
apiVersion: v1
kind: ServiceAccount
metadata:
  labels:
    app: metallb
  name: speaker
  namespace: metallb-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  labels:
    app: metallb
  name: metallb-system:controller
rules:
- apiGroups:
  - ''
  resources:
  - services
  verbs:
  - get
  - list
  - watch
  - update
- apiGroups:
  - ''
  resources:
  - services/status
  verbs:
  - update
- apiGroups:
  - ''
  resources:
  - events
  verbs:
  - create
  - patch
- apiGroups:
  - policy
  resourceNames:
  - controller
  resources:
  - podsecuritypolicies
  verbs:
  - use
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
  labels:
    app: metallb
  name: metallb-system:speaker
rules:
- apiGroups:
  - ''
  resources:
  - services
  - endpoints
  - nodes
  verbs:
  - get
  - list
  - watch
- apiGroups:
  - ''
  resources:
  - events
  verbs:
  - create
  - patch
- apiGroups:
  - policy
  resourceNames:
  - speaker
  resources:
  - podsecuritypolicies
  verbs:
  - use
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  labels:
    app: metallb
  name: config-watcher
  namespace: metallb-system
rules:
- apiGroups:
  - ''
  resources:
  - configmaps
  verbs:
  - get
  - list
  - watch
---
apiVersion: rbac.authorization.k8s.io/v1
kind: Role
metadata:
  labels:
    app: metallb
  name: pod-lister
  namespace: metallb-system
rules:
- apiGroups:
  - ''
  resources:
  - pods
  verbs:
  - list
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  labels:
    app: metallb
  name: metallb-system:controller
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: metallb-system:controller
subjects:
- kind: ServiceAccount
  name: controller
  namespace: metallb-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  labels:
    app: metallb
  name: metallb-system:speaker
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: metallb-system:speaker
subjects:
- kind: ServiceAccount
  name: speaker
  namespace: metallb-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  labels:
    app: metallb
  name: config-watcher
  namespace: metallb-system
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: config-watcher
subjects:
- kind: ServiceAccount
  name: controller
- kind: ServiceAccount
  name: speaker
---
apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  labels:
    app: metallb
  name: pod-lister
  namespace: metallb-system
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: pod-lister
subjects:
- kind: ServiceAccount
  name: speaker
---
apiVersion: apps/v1
kind: DaemonSet
metadata:
  labels:
    app: metallb
    component: speaker
  name: speaker
  namespace: metallb-system
spec:
  selector:
    matchLabels:
      app: metallb
      component: speaker
  template:
    metadata:
      annotations:
        prometheus.io/port: '7472'
        prometheus.io/scrape: 'true'
      labels:
        app: metallb
        component: speaker
    spec:
      containers:
      - args:
        - --port=7472
        - --config=config
        env:
        - name: METALLB_NODE_NAME
          valueFrom:
            fieldRef:
              fieldPath: spec.nodeName
        - name: METALLB_HOST
          valueFrom:
            fieldRef:
              fieldPath: status.hostIP
        - name: METALLB_ML_BIND_ADDR
          valueFrom:
            fieldRef:
              fieldPath: status.podIP
        - name: METALLB_ML_LABELS
          value: "app=metallb,component=speaker"
        - name: METALLB_ML_NAMESPACE
          valueFrom:
            fieldRef:
              fieldPath: metadata.namespace
        - name: METALLB_ML_SECRET_KEY
          valueFrom:
            secretKeyRef:
              name: memberlist
              key: secretkey
        image: docker.mirrors.ustc.edu.cn/metallb/speaker:v0.9.3
        imagePullPolicy: IfNotPresent
        name: speaker
        ports:
        - containerPort: 7472
          name: monitoring
        resources:
          limits:
            cpu: 100m
            memory: 100Mi
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            add:
            - NET_ADMIN
            - NET_RAW
            - SYS_ADMIN
            drop:
            - ALL
          readOnlyRootFilesystem: true
      hostNetwork: true
      nodeSelector:
        beta.kubernetes.io/os: linux
      serviceAccountName: speaker
      terminationGracePeriodSeconds: 2
      tolerations:
      - effect: NoSchedule
        key: node-role.kubernetes.io/master
---
apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: metallb
    component: controller
  name: controller
  namespace: metallb-system
spec:
  revisionHistoryLimit: 3
  selector:
    matchLabels:
      app: metallb
      component: controller
  template:
    metadata:
      annotations:
        prometheus.io/port: '7472'
        prometheus.io/scrape: 'true'
      labels:
        app: metallb
        component: controller
    spec:
      containers:
      - args:
        - --port=7472
        - --config=config
        image: docker.mirrors.ustc.edu.cn/metallb/controller:v0.9.3
        imagePullPolicy: IfNotPresent
        name: controller
        ports:
        - containerPort: 7472
          name: monitoring
        resources:
          limits:
            cpu: 100m
            memory: 100Mi
        securityContext:
          allowPrivilegeEscalation: false
          capabilities:
            drop:
            - all
          readOnlyRootFilesystem: true
      nodeSelector:
        beta.kubernetes.io/os: linux
      securityContext:
        runAsNonRoot: true
        runAsUser: 65534
      serviceAccountName: controller
      terminationGracePeriodSeconds: 0

创建memberlist

kubectl create secret generic -n metallb-system memberlist --from-literal=secretkey="$(openssl rand -base64 128)"

部署完成以后，没有任何配置，还不能直接使用，接下来我们配置Layer2模式

kind: ConfigMap
apiVersion: v1
metadata:
  name: config
  namespace: metallb-system
data:
  config: |
    address-pools:
    - name: default
      protocol: layer2
      addresses:
      - 10.26.32.220-10.26.32.225

Layer2原理示意图

Layer 2 模式的优缺点：

• Layer 2 模式更为通用，不需要用户有额外的设备；

• Layer 2 模式下存在单点问题，服务的所有入口流量经由单点，其网络带宽可能成为瓶颈；

• 由于 Layer 2 模式需要 ARP/NDP 客户端配合，当故障转移发生时，MetalLB 会发送 ARP 包来宣告 MAC 地址和 IP 映射关系的变化，地址分配略为繁琐。

测试

部署一个测试服务，这里以nginx为例

---
apiVersion: apps/v1
kind: StatefulSet
metadata:
  name: demo001
  namespace: default
  labels:
    app: demo001
spec:
  serviceName: demo001
  replicas: 1
  selector:
    matchLabels:
      app: demo001
  template:
    metadata:
      labels:
        app: demo001
    spec:
      terminationGracePeriodSeconds: 180
      nodeSelector:
        kubernetes.io/hostname: 10.26.32.169
      containers:
      - name: demo001
        image: nginx
        imagePullPolicy: IfNotPresent
        ports:
        - containerPort: 80
          name: port
        volumeMounts:
        - name: volume
          mountPath: /var/www/html
  volumeClaimTemplates:
  - metadata:
      name: volume
    spec:
      accessModes: ["ReadWriteOnce"]
      storageClassName: rook-ceph
      resources:
        requests:
          storage: 1Gi
---
kind: Service
apiVersion: v1
metadata:
  name: nginx-svc-lb
  namespace: default
spec:
  ports:
  - port: 80
    targetPort: 80
  selector:
    app: demo001
  type: LoadBalancer

• 部署结果

NAME                          TYPE           CLUSTER-IP      EXTERNAL-IP       PORT(S)                       AGE
nginx-svc-lb              LoadBalancer   10.96.92.102    192.168.100.220   80:32545/TCP                92m

然后现在就可以直接用EXTERNAL-IP: 192.168.100.220访问到服务了

# telnet  192.168.100.220 80
Trying 192.168.100.220...
Connected to 192.168.100.220.
Escape character is '^]'.