7.k8s.調度器scheduler 親和性、污點

#k8s. 調度器scheduler 親和性、污點

默認調度過程：預選 Predicates (過濾節點) --> 優選 Priorities(優先級排序) --> 優先級最高節點
實際使用，根據需求控制Pod調度，需要用到如下:
指定節點、nodeAffinity(節點親和性)、podAffinity(pod 親和性)、 podAntiAffinity(pod 反親和性)

#指定調度節點

# Pod.spec.nodeName 強制匹配，直接指定Node 節點，跳過 Scheduler 調度策略

#node-name-demo.yaml
apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: demo-nodename
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: demo1
    spec:
      nodeName: node03  #指定Node節點
      containers:
      - name: demo1
        image: alivv/nginx:node
        ports:
        - containerPort: 80

#部署
kubectl apply -f node-name-demo.yaml
#查看pod全部在node03上 (node03節點不存在會一直處於pending)
kubectl get pod -o wide
#刪除
kubectl delete -f node-name-demo.yaml

# Pod.spec.nodeSelector 強制約束，調度策略匹配 label，調度 Pod 到目標節點

#node-selector-demo.yaml
apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: demo-node-selector
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: demo1
    spec:
      nodeSelector:
        test1: node  #匹配lable test1=node
      containers:
      - name: demo1
        image: alivv/nginx:node
        ports:
        - containerPort: 80

#部署
kubectl apply -f node-selector-demo.yaml

#查看pod處於pending
kubectl get pod -o wide
#給node02節點添加lable
kubectl label nodes node02 test1=node
kubectl get nodes --show-labels
#再次查看pod在node02節點
kubectl get pod -o wide

#刪除
kubectl delete -f node-selector-demo.yaml
kubectl label nodes node02 test1-

親和性調度

親和性調度可以分成軟策略和硬策略兩種方式

• preferredDuringSchedulingIgnoredDuringExecution：軟策略，沒滿足條件就忽略，Pod可以啟動
• requiredDuringSchedulingIgnoredDuringExecution：硬策略，沒滿足條件就等待，Pod處於Pending

操作符

• In：label 的值在某個列表中
• NotIn：label 的值不在某個列表中
• Gt：label 的值大於某個值
• Lt：label 的值小於某個值
• Exists：某個 label 存在
• DoesNotExist：某個 label 不存在

#節點親和性 pod.spec.nodeAffinity

#node-affinity-demo.yaml
apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: node-affinity
  labels:
    app: affinity
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: alivv/nginx:node
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  #硬策略,不在node01節點
            nodeSelectorTerms:
            - matchExpressions:
              - key: kubernetes.io/hostname
                operator: NotIn
                values:
                - node01
          preferredDuringSchedulingIgnoredDuringExecution:  #軟策略,優先匹配test2=node
          - weight: 1
            preference:
              matchExpressions:
              - key: test2
                operator: In
                values:
                - node

#給node03節點添加lable
kubectl label nodes node03 test2=node
kubectl get nodes --show-labels

#部署
kubectl apply -f node-affinity-demo.yaml

#查看pod
kubectl get pod -o wide

#刪除
kubectl delete -f node-affinity-demo.yaml
kubectl label nodes node03 test2-

#Pod親和性pod.spec.affinity.podAffinity/podAntiAffinity

podAffinity Pod親和性，解決 pod 部署在同一個拓撲域 、或同一個節點
podAntiAffinity Pod反親和性，避開Pod部署在一起

#pod-affinity-demo.yaml
apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: pod-affinity-demo
  labels:
    app: pod-affinity
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: pod-affinity
    spec:
      containers:
      - name: nginx
        image: alivv/nginx:node
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        podAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  #硬策略
          - labelSelector:  #匹配Pod有 app=demo1
              matchExpressions:
              - key: app
                operator: In
                values:
                - demo1
            topologyKey: kubernetes.io/hostname

#部署
kubectl apply -f pod-affinity-demo.yaml

#查看pod全部處於Pending 因為沒標簽app=demo1的Pod
kubectl get pod -o wide

#部署上面的node-name-demo.yaml
kubectl apply -f  node-name-demo.yaml
#再次查看pod全部在node03節點
kubectl get pod -o wide

#Pod反親和性測試
#改podAffinity為podAntiAffinity
sed -i 's/podAffinity/podAntiAffinity/' pod-affinity-demo.yaml
kubectl apply -f pod-affinity-demo.yaml

#查看node03節點移除pod-affinity-demo
kubectl get pod -o wide

#刪除
kubectl delete -f pod-affinity-demo.yaml
kubectl delete -f node-name-demo.yaml

# 污點taints與容忍tolerations

節點標記為 Taints ，除非 pod可以容忍污點節點，否則該 Taints 節點不會被調度pod
kubeadm安裝k8s，默認master節點會添加NoSchedule 類型污點

污點設置 kubectl taint nodes node-name key=value:effect
key 和 value 為污點標簽， value 可以為空，effect 描述污點的作用，effect 支持如下三個選項：

• NoSchedule 不會將 Pod 調度到有污點的 Node
• PreferNoSchedule 避免將 Pod 調度到有污點的 Node
• NoExecute 不會將 Pod 調度到有污點的 Node ，將已經存在的 Pod 驅逐出去

#給node03添加污點
kubectl taint nodes node03 test-taint=node:NoSchedule
#查看
kubectl describe node node03 |grep Taints

容忍 pod.spec.tolerations

#pod-tolerations-demo.yaml
apiVersion: apps/v1beta1
kind: Deployment
metadata:
  name: pod-tolerations-demo
  labels:
    app: pod-tolerations
spec:
  replicas: 3
  template:
    metadata:
      labels:
        app: pod-tolerations
    spec:
      containers:
      - name: nginx
        image: alivv/nginx:node
        ports:
        - containerPort: 80
          name: http

#創建Pod
kubectl apply -f pod-tolerations-demo.yaml

#查看Pod，node03節點有污點將不會創建Pod
kubectl get pod -o wide

#文件pod-tolerations-demo.yaml追加容忍污點配置
echo '#容忍污點
      tolerations:
      - key: "test-taint"
        #value: "node"
        operator: "Exists"  #忽略value值
        effect: "NoSchedule"
'>>pod-tolerations-demo.yaml
cat pod-tolerations-demo.yaml

#更新
kubectl apply -f pod-tolerations-demo.yaml
#Pod擴容
kubectl scale deployment pod-tolerations-demo --replicas 5

#再次查看Pod，Node03有Pod
kubectl get pod -o wide

#刪除Pod
kubectl delete -f pod-tolerations-demo.yaml
#刪除污點
kubectl taint nodes node03 test-taint-

#不指定 key 值時，容忍所有的污點 key

tolerations:
- operator: "Exist

#不指定 effect 值時，表示容忍所有的污點作用

tolerations:
- key: "key"
  operator: "Exist"

#避免資源浪費，設置master運行Pod，可以如下配置：

kubectl taint nodes --all  node-role.kubernetes.io/master- #先刪除默認污點
kubectl taint nodes Node-Name node-role.kubernetes.io/master=:PreferNoSchedule

---

Blog地址 https://www.cnblogs.com/elvi/p/11755828.html
本文git地址 https://gitee.com/almi/k8s/tree/master/notes