浅谈Kubernetes的节点选择，亲和度使用和污点的使用

一、nodeSelector：节点的选择是一种硬性的要求，主动性的选择。在不考虑污点的情况下，直接指定Pod部署到哪个节点中。判断的依据是通过node节点的标签来进行判断

示例

kubectl label nodes [节点名称] [标签名称]=[标签值]

apiVersion: v1
kind: Pod
metadata:
  name: nodetest
  labels:
    app: nodetest
spec:
  restartPolicy: Always
  nodeSelector:
    disktype: ssd
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
    resources:
      requests:
        cpu: 500m
        memory: 64Mi
      limits:
        cpu: 600m
        memory: 128Mi

二、nodeAffinity：亲和度，这里分硬亲和度和软亲和度。硬亲和度类似于nodeSelector一般，必须key和vvalue对的上才行。而软亲和度是尽可能的匹配。

示例

硬亲和度

apiVersion: v1
kind: Pod
metadata:
  name: nodetest
  labels:
    app: nodetest
spec:
  restartPolicy: Always
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: disktype
            operator: In
            values:
            - ssd 
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
    resources:
      requests:
        cpu: 500m
        memory: 64Mi
      limits:
        cpu: 600m
        memory: 128Mi

软亲和度

apiVersion: v1
kind: Pod
metadata:
  name: nodetest
  labels:
    app: nodetest
spec:
  restartPolicy: Always
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 1
        preference:
          matchExpressions:
          - key: disktype
            operator: In
            values:
            - ssd  
            - ssh
            - abc 
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
    resources:
      requests:
        cpu: 500m
        memory: 64Mi
      limits:
        cpu: 600m
        memory: 128Mi

三、污点，我的理解是一种被动的标记。节点选择，亲和度和污点配合起来，更高效的实现pod的合理分布

设置污点

kubectl taint nodes [节点名称] key=value:[effect]

effect：NoSchedule 不允许调度、PreferNoSchedule 尽量不调度、NoExecute 不允许调度并且驱逐节点内的Pod

查看节点的污点

kubectl describe nodes | grep -i taint

污点如果设置到了node上。那么就会产生对应的效果。实际情况下，设置污点是为了更好的利用node资源。那么对应就有一个容忍度来做这个事情。容忍度规则内，那么污点是允许通过的。

示例

apiVersion: v1
kind: Pod
metadata:
  name: nodetest
  labels:
    app: nodetest
spec:
  restartPolicy: Always
  tolerations:
  - key: "disktype"
    operator: "Exists"
    effect: "NoSchedule"
  #tolerations:
  #- key: "disktype"
  #  operator: "Equal"
  #  value: "ssh"
  #  effect: "NoSchedule"

  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent

容忍度的选项有必须符合和包含。必须符合是必须key=value，包含存在的情况是存在这个key即可