vlambda博客
学习文章列表

写在前面

• 学习 K8s,这块官方文档的翻译的有点生涩，不太好明白。所以整理下
• 博文内容涉及
• Service Account的简述
• Service Account的Demo
• 官方文档:https://kubernetes.io/zh/docs/tasks/configure-pod-container/configure-service-account/

「 真正的坚持归于平静，靠的是温和的发力，而不是时时刻刻的刺激。」

学习环境

┌──[[email protected]]-[~]
└─$kubectl get nodes
NAME                         STATUS     ROLES                  AGE    VERSION
vms81.liruilongs.github.io   Ready      control-plane,master   134d   v1.22.2
vms82.liruilongs.github.io   Ready      <none>                 134d   v1.22.2
vms83.liruilongs.github.io   NotReady   <none>                 134d   v1.22.2
┌──[[email protected]]-[~]
└─$

Service Account 是什么？

学习Service Account之前，简单介绍下K8s的安全体系，K8s中通过一系列机制来实现集群的安全控制，其中包括API Server的认证和授(鉴)权，关于认证和授(鉴)权，感兴趣小伙伴可以看看之前的博文，我们这里简单介绍下

「关于授(鉴)权，现在用的比较多的是RBAC(Role-Based Access Control,基于角色的访问控制)的方式」

RBAC在Kubernetes的1.5版本中引入，在1.6版本时升级为Beta版本，在1.8版本时升级为GA。现在作为kubeadm安装方式的默认选项，相对于其他访问控制方式，RBAC对集群中的资源和非资源权限均有完整的覆盖。整个RBAC完全由(Role,ClusterRole,RoleBinding,ClusterRoleBinding)API对象完成，同其他API对象一样，可以用kubectl或API进行操作。可以在运行时进行调整，无须重新启动 API Server。

K8s的授权策略设置通过通过API Server的启动参数"--authorization-mode"设置。除了RBAC外，授权策略还包括：

「关于认证机制，在K8s的认证中，如果按照集群内外认证分的话，分为集群外认证和集群内认证：」

集群外认证一般三种，也可以理解为通过kubectl或者编程语言编写的客户端API访问：

• HTTP Token认证：通过一个 Token来识别合法用户。
• HTTPS 证书认证：基于CA根证书签名的双向数字证书认证方式(Kubeconfig文件)
• HTTP Base认证：通过用户名+密码的方式认证(用户账户)，这个只有 1.19之前的版本适用，之后的版本不在支持

集群内的认证也就是我们今天要讲的：Service Account对象,也叫服务账户

所以说Service Account它并不是给Kubernetes集群的用户(系统管理员、运维人员)用的,而是给运行在K8s上的Pod里的进程用的,为Pod里的进程提供认证。

比如我们要编写一个类似kubectl一样的K8s的管理工具，如一些面板工具(kubernetes-dashboard)，而且这个工具是运行在我们的K8s环境里的，那么这个时候，我们如何给这个工具访问集群做认证授权，就要用到Service Account，简写为sa,所以我们一般直接叫sa，或者服务账户

当我们创建任何一个Pod的时候，必须要有sa，否则会创建失败，如果没有显示的指定对应的sa，即服务账户，Pod会默认使用当前的命令空间的default服务账户(每个命名空间都有一个名为 default 的sa资源。)

这里要说明的是每个sa服务账户都会生成一个secret，这个secret里面包含一个token凭证。所以说sa实际认证是通过token实现的认证。(token)

┌──[[email protected]]-[~]
└─$kubectl get sa default
NAME      SECRETS   AGE
default   1         67d
┌──[[email protected]]-[~]
└─$kubectl run podcommon --image=nginx --image-pull-policy=IfNotPresent --labels="name=liruilong" --env="name=liruilong"
pod/podcommon created
┌──[[email protected]]-[~]
└─$kubectl get pods podcommon  -o yaml | grep serviceAccount
  serviceAccount: default
  serviceAccountName: default
      - serviceAccountToken:
┌──[[email protected]]-[~]
└─$

我们可以使用自动挂载给Pod的default服务账户 token访问 API，但是前提是需要给default授权，对于RBAC的方式来讲，需要给角色授权，然后绑定角色。

在 1.6 以上版本中，可以通过在sa上设置automountServiceAccountToken: false来实现不给服务账号自动挂载 API token：

apiVersion: v1
kind: ServiceAccount
metadata:
  name: build-robot
automountServiceAccountToken: false
...

在 1.6 以上版本中，你也可以选择不给特定 Pod 自动挂载 API token：

apiVersion: v1
kind: Pod
metadata:
  name: my-pod
spec:
  serviceAccountName: build-robot
  automountServiceAccountToken: false
  ...

如果 Pod 和服务账户都指定了automountServiceAccountToken值，则 Pod 的 spec 优先于服务帐户。

下面看一下kubernetes-dashboard对sa的应用,下面是一个已经部署好的dashboard

关于kubernetes-dashboard是K8s官网提供的Kubernetes的Web UI网页管理工具，可提供部署应用、资源对象管理、容器日志查询、系统监控等常用的集群管理功能。

┌──[[email protected]]-[~]
└─$kubectl get pods -o wide
NAME                                         READY   STATUS    RESTARTS        AGE   IP             NODE
         NOMINATED NODE   READINESS GATES
dashboard-metrics-scraper-669c88c9d9-2qp62   1/1     Running   8 (7d11h ago)   61d   10.244.88.83   vms81.liruilongs.github.io   <none>           <none>
kubernetes-dashboard-5d66bcd8fd-l22jm        1/1     Running   13 (7d3h ago)   61d   10.244.88.80   vms81.liruilongs.github.io   <none>           <none>
┌──[[email protected]]-[~]
└─$
┌──[[email protected]]-[~]
└─$kubectl get svc  -n kubernetes-dashboard
NAME                        TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)         AGE
dashboard-metrics-scraper   ClusterIP   10.109.92.159   <none>        8000/TCP        67d
kubernetes-dashboard        NodePort    10.106.48.37    <none>        443:32360/TCP   67d
┌──[[email protected]]-[~]
└─$

上面是一个我们之前部署好的面板工具，在部署的过程中，我们要主动创建一个sa(kubernetes-dashboard)，并且为这个sa授权，而后，我们的这个面板工具才具有管理K8s集群的能力

创建sa的资源文件

apiVersion: v1
kind: ServiceAccount
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard

查看kubernetes-dashboard sa，可以看到对应的token

┌──[[email protected]]-[~]
└─$kubectl get sa
NAME                   SECRETS   AGE
......
kubernetes-dashboard   1         67d
┌──[[email protected]]-[~]
└─$kubectl get secrets
NAME                               TYPE                                  DATA   AGE
.........
kubernetes-dashboard-token-wnqqg   kubernetes.io/service-account-token   3      67d
┌──[[email protected]]-[~]
└─$kubectl describe  secrets kubernetes-dashboard-token-wnqqg
Name:         kubernetes-dashboard-token-wnqqg
Namespace:    kubernetes-dashboard
Labels:       <none>
Annotations:  kubernetes.io/service-account.name: kubernetes-dashboard
              kubernetes.io/service-account.uid: 8e209de5-14a0-4dd5-bd19-2264170531f5

Type:  kubernetes.io/service-account-token

Data
====
token:      eyJhbGciOiJSUzI1NiIsImtpZCI6ImF2MmJVZ3d6M21JRC1BZUwwaHlDdzZHSGNyaVJON1BkUHF6MlhPV2NfX00ifQ.eyJpc3MiOiJrdWJlcm5ldGVzL3NlcnZpY2VhY2NvdW50Iiwia3ViZXJuZXRlcy5pby9zZXJ2aWNlYWNjb3VudC9uYW1lc3BhY2UiOiJrdWJlcm5ldGVzLWRhc2hib2FyZCIsImt1YmVybmV0ZXMuaW8vc2VydmljZWFjY291bnQvc2VjcmV0Lm5hbWUiOiJrdW....
ca.crt:     1099 bytes
namespace:  20 bytes
┌──[[email protected]]-[~]
└─$

然后对sa授权，一般通过RBAC的方式.创建角色

kind: ClusterRole
apiVersion: rbac.authorization.k8s.io/v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
rules:
  # Allow Metrics Scraper to get metrics from the Metrics server
  - apiGroups: ["metrics.k8s.io"]
    resources: ["pods", "nodes"]
    verbs: ["get", "list", "watch"]

然后绑定角色到sa

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: kubernetes-dashboard
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: kubernetes-dashboard
subjects:
  - kind: ServiceAccount  
    name: kubernetes-dashboard
    namespace: kubernetes-dashboard

然后pod通过serviceAccount和serviceAccountName来绑定sa，当然这两个参数指定一个就可以了。

┌──[[email protected]]-[~]
└─$kubectl get  pod kubernetes-dashboard-5d66bcd8fd-l22jm -o yaml | grep -C 5  serviceAccount
  preemptionPolicy: PreemptLowerPriority
  priority: 0
  restartPolicy: Always
  schedulerName: default-scheduler
  securityContext: {}
  serviceAccount: kubernetes-dashboard
  serviceAccountName: kubernetes-dashboard
  terminationGracePeriodSeconds: 30

通过yaml文件我们可以看到，值sa为kubernetes-dashboard,当然在资源文件中，是在Deployment和Servcie中指定，

如果sa的automountServiceAccountToken或Pod的automountServiceAccountToken都未显式设置为 false，那么会为对应的 Pod 创建一个 volume，在其中包含用来访问 API 的令牌。

如果为sa对应的token创建了卷，则为 Pod 中的每个容器添加一个 volumeSource，挂载在其 /var/run/secrets/kubernetes.io/serviceaccount 目录下。

┌──[[email protected]]-[~]
└─$kubectl get  pod kubernetes-dashboard-5d66bcd8fd-l22jm -o yaml | grep -C 20 -i  serviceAccount
...........
    volumeMounts:
...........
    - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
      name: kube-api-access-8jlj7
      readOnly: true
.........
  serviceAccount: kubernetes-dashboard
  serviceAccountName: kubernetes-dashboard
.........
  volumes:
..........
  - name: kube-api-access-8jlj7
    projected:
      defaultMode: 420
      sources:
      - serviceAccountToken:
          expirationSeconds: 3607
          path: token
........
┌──[[email protected]]-[~]
└─$

通过配置文件可以看到,token通过卷的方式挂载到了容器里的/var/run/secrets/kubernetes.io/serviceaccount 目录，但是需要注意的是，这个token和sa对应的token在1.20版本之后进行了处理，不一样，在之前的版本中是一样的。

Service Account Demo

创建一个sa

┌──[[email protected]]-[~]
└─$kubectl create sa  sa-demo
serviceaccount/sa-demo created

查看对应的secret和token

┌──[[email protected]]-[~]
└─$kubectl get secrets sa-demo-token-pdrs8
NAME                  TYPE                                  DATA   AGE
sa-demo-token-pdrs8   kubernetes.io/service-account-token   3      43s
┌──[[email protected]]-[~]
└─$kubectl describe secrets sa-demo-token-pdrs8
Name:         sa-demo-token-pdrs8
Namespace:    kubernetes-dashboard
Labels:       <none>
Annotations:  kubernetes.io/service-account.name: sa-demo
              kubernetes.io/service-account.uid: 7003de88-803a-4dae-a6e3-d647d0517c92

Type:  kubernetes.io/service-account-token

Data
====
ca.crt:     1099 bytes
namespace:  20 bytes
token:      eyJhbGciOiJSUzI1NiIsImtpZCI6ImF2MmJVZ3d6M21JRC1BZUwwaHlDdzZHSGNyaVJON1BkUHF6MlhPV2NfX00ifQ.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.sVAmtpfqFREjUCd9bkQvMuHpasXOcKYLJvVsJLLe6ufP4zs8ZVt6HqH4ylsxbmwtibNXBV9hVNEU_2X3T2enOjOSYuiyaEP4BifDQN7DmZbu2uXQCBglixaNB7ZIIPX_oQsW0ndBNonVqMSMm-ZItYDzLo-QTOxTxc5OQZ3zSBJqITAvWFlshWA7mKntNmWw6m5KunjhYZs14Lpa-NhknYS9G6ur8SKY4XdE44hzQhD7h4y01ZezZGR3IdGd3HktA5dWYTRXXr9H00odey2YtGfj40Vql3rMrdMPJOFbAozjyaWxhmSpjHVGcbXawai8znKPCdGlW4l2aRmbghovsw
┌──[[email protected]]-[~]
└─$

编写pod资源文件，指定sa为刚才创建的sa

┌──[[email protected]]-[~]
└─$kubectl run pod-sa --image=nginx --image-pull-policy=IfNotPresent --dry-run=client  -o yaml > pod-sa.yaml
┌──[[email protected]]-[~]
└─$vim pod-sa.yaml
┌──[[email protected]]-[~]
└─$cat pod-sa.yaml
apiVersion: v1
kind: Pod
metadata:
  creationTimestamp: null
  labels:
    run: pod-demo
  name: pod-demo
spec:
  serviceAccount: sa-demo
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-demo
    resources: {}
  dnsPolicy: ClusterFirst
  restartPolicy: Always
status: {}
┌──[[email protected]]-[~]
└─$

查看创建的pod

┌──[[email protected]]-[~]
└─$kubectl apply -f pod-sa.yaml
pod/pod-demo created
┌──[[email protected]]-[~]
└─$kubectl get pods pod-demo
NAME       READY   STATUS    RESTARTS   AGE
pod-demo   1/1     Running   0          95s
┌──[[email protected]]-[~]
└─$

对于deplay的sa修改可以直接通过set的方式设置，时间关系这里不多讲啦，文末的资源文件中有demo

下面我们来看一道Service Account相关习题,这是某一期CKA认证的一道考题

创建一个名为deployment-clusterrole且仅允许创建以下资源类型的新ClusterRole：

• Deployment
• StatefulSet
• DaemonSet

在现有的 namespace app-team 中创建一个名为cicd-token的新 ServiceAccount。

限于 namespace app-team 中，将新的ClusterRole deployment-clusterrole绑定到新的 ServiceAccount cicd-token。

题目很简单，一般的生产我们也会涉及，指定权限创建一个集群角色，然后把这个集群角色绑定到一个新建的sa上。

# 创建集群角色
kubectl create clusterrole deployment-clusterrole --verb=create --resource=deployments,statefulsets,daemonsets
# 创建服务账户
kubectl -n app-team create serviceaccount cicd-token
# 绑定角色到服务账户
kubectl -n app-team create rolebinding cicd-token-rolebinding --clusterrole=deployment-clusterrole --serviceaccount=app-team:cicd-token

对sa学习，感觉kubernetes-dashboard的是一个很好的Demo。这里把面板的资源文件贴出来，感兴趣小伙伴可以研究下

# Copyright 2017 The Kubernetes Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

apiVersion: v1
kind: Namespace
metadata:
  name: kubernetes-dashboard

---

apiVersion: v1
kind: ServiceAccount
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard

---

kind: Service
apiVersion: v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard
spec:
  ports:
    - port: 443
      targetPort: 8443
  selector:
    k8s-app: kubernetes-dashboard

---

apiVersion: v1
kind: Secret
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard-certs
  namespace: kubernetes-dashboard
type: Opaque

---

apiVersion: v1
kind: Secret
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard-csrf
  namespace: kubernetes-dashboard
type: Opaque
data:
  csrf: ""

---

apiVersion: v1
kind: Secret
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard-key-holder
  namespace: kubernetes-dashboard
type: Opaque

---

kind: ConfigMap
apiVersion: v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard-settings
  namespace: kubernetes-dashboard

---

kind: Role
apiVersion: rbac.authorization.k8s.io/v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard
rules:
  # Allow Dashboard to get, update and delete Dashboard exclusive secrets.
  - apiGroups: [""]
    resources: ["secrets"]
    resourceNames: ["kubernetes-dashboard-key-holder", "kubernetes-dashboard-certs", "kubernetes-dashboard-csrf"]
    verbs: ["get", "update", "delete"]
    # Allow Dashboard to get and update 'kubernetes-dashboard-settings' config map.
  - apiGroups: [""]
    resources: ["configmaps"]
    resourceNames: ["kubernetes-dashboard-settings"]
    verbs: ["get", "update"]
    # Allow Dashboard to get metrics.
  - apiGroups: [""]
    resources: ["services"]
    resourceNames: ["heapster", "dashboard-metrics-scraper"]
    verbs: ["proxy"]
  - apiGroups: [""]
    resources: ["services/proxy"]
    resourceNames: ["heapster", "http:heapster:", "https:heapster:", "dashboard-metrics-scraper", "http:dashboard-metrics-scraper"]
    verbs: ["get"]

---

kind: ClusterRole
apiVersion: rbac.authorization.k8s.io/v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
rules:
  # Allow Metrics Scraper to get metrics from the Metrics server
  - apiGroups: ["metrics.k8s.io"]
    resources: ["pods", "nodes"]
    verbs: ["get", "list", "watch"]

---

apiVersion: rbac.authorization.k8s.io/v1
kind: RoleBinding
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: Role
  name: kubernetes-dashboard
subjects:
  - kind: ServiceAccount
    name: kubernetes-dashboard
    namespace: kubernetes-dashboard

---

apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
  name: kubernetes-dashboard
roleRef:
  apiGroup: rbac.authorization.k8s.io
  kind: ClusterRole
  name: kubernetes-dashboard
subjects:
  - kind: ServiceAccount
    name: kubernetes-dashboard
    namespace: kubernetes-dashboard

---

kind: Deployment
apiVersion: apps/v1
metadata:
  labels:
    k8s-app: kubernetes-dashboard
  name: kubernetes-dashboard
  namespace: kubernetes-dashboard
spec:
  replicas: 1
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      k8s-app: kubernetes-dashboard
  template:
    metadata:
      labels:
        k8s-app: kubernetes-dashboard
    spec:
      containers:
        - name: kubernetes-dashboard
          #image: kubernetesui/dashboard:v2.0.0-beta8
          image: registry.cn-hangzhou.aliyuncs.com/kube-iamges/dashboard:v2.0.0-beta8
          #imagePullPolicy: Always
          imagePullPolicy: IfNotPresent
          ports:
            - containerPort: 8443
              protocol: TCP
          args:
            - --auto-generate-certificates
            - --namespace=kubernetes-dashboard
            # Uncomment the following line to manually specify Kubernetes API server Host
            # If not specified, Dashboard will attempt to auto discover the API server and connect
            # to it. Uncomment only if the default does not work.
            # - --apiserver-host=http://my-address:port
          volumeMounts:
            - name: kubernetes-dashboard-certs
              mountPath: /certs
              # Create on-disk volume to store exec logs
            - mountPath: /tmp
              name: tmp-volume
          livenessProbe:
            httpGet:
              scheme: HTTPS
              path: /
              port: 8443
            initialDelaySeconds: 30
            timeoutSeconds: 30
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            runAsUser: 1001
            runAsGroup: 2001
      volumes:
        - name: kubernetes-dashboard-certs
          secret:
            secretName: kubernetes-dashboard-certs
        - name: tmp-volume
          emptyDir: {}
      serviceAccountName: kubernetes-dashboard
      nodeSelector:
        "kubernetes.io/os": linux
      # Comment the following tolerations if Dashboard must not be deployed on master
      tolerations:
        - key: node-role.kubernetes.io/master
          effect: NoSchedule

---

kind: Service
apiVersion: v1
metadata:
  labels:
    k8s-app: dashboard-metrics-scraper
  name: dashboard-metrics-scraper
  namespace: kubernetes-dashboard
spec:
  ports:
    - port: 8000
      targetPort: 8000
  selector:
    k8s-app: dashboard-metrics-scraper

---

kind: Deployment
apiVersion: apps/v1
metadata:
  labels:
    k8s-app: dashboard-metrics-scraper
  name: dashboard-metrics-scraper
  namespace: kubernetes-dashboard
spec:
  replicas: 1
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      k8s-app: dashboard-metrics-scraper
  template:
    metadata:
      labels:
        k8s-app: dashboard-metrics-scraper
      annotations:
        seccompProfile: 'runtime/default'
    spec:
      containers:
        - name: dashboard-metrics-scraper
          #image: kubernetesui/metrics-scraper:v1.0.1
          image: registry.cn-hangzhou.aliyuncs.com/kube-iamges/metrics-scraper:v1.0.1
          imagePullPolicy: IfNotPresent
          ports:
            - containerPort: 8000
              protocol: TCP
          livenessProbe:
            httpGet:
              scheme: HTTP
              path: /
              port: 8000
            initialDelaySeconds: 30
            timeoutSeconds: 30
          volumeMounts:
          - mountPath: /tmp
            name: tmp-volume
          securityContext:
            allowPrivilegeEscalation: false
            readOnlyRootFilesystem: true
            runAsUser: 1001
            runAsGroup: 2001
      serviceAccountName: kubernetes-dashboard
      nodeSelector:
        "kubernetes.io/os": linux
      # Comment the following tolerations if Dashboard must not be deployed on master
      tolerations:
        - key: node-role.kubernetes.io/master
          effect: NoSchedule
      volumes:
        - name: tmp-volume
          emptyDir: {}