Pod Scheduling Strategy: Affinity, Stain and Stain Tolerance

节点亲和性

nodeSelector

nodeSelector用于将Pod调度到匹配Label的节点上,如果没有匹配的标签会调度失败.

作用：

• 约束Pod到特定节点运行;
• 完全匹配节点标签.

应用场景：

• 专用节点：Nodes are grouped and managed according to business lines;
• Match special hardware：Some nodes are configured with SSD硬盘、GPU.

---

示例：确保Pod分配到具有SSD硬盘的节点上

1. 给节点打标签

#kubectl label nodes <node名称> <label-key>=<label-value>
kubectl label nodes k8s-node1 disktype=ssd
#验证
kubectl get nodes --show-labels

2. 添加nodeSelector字段到Pod配置中

apiVersion: v1
kind: Pod
metadata:
  name: nginx
  labels:
    env: test
spec:
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
  nodeSelector:
    disktype: ssd

3. 验证

kubectl get pods -o wide

删除节点标签：

kubectl label node k8s-node1 <label-key>-
#验证
kubectl get pods -o wide

nodeAffinity

节点亲和性,类似于nodeSelector,可以根据节点上的标签来约束Pod可以调度到哪些节点.

相比nodeSelector：

• 匹配有更多的逻辑组合,不只是字符串的完全相等,支持in、notin、exists、doesnotexist、gt、lt操作符.
• 调度分为软策略和硬策略,而不是硬性要求：
  • requiredDuringSchedulingIgnoredDuringExecution：硬策略,必须满足;
  • preferredDuringSchedulingIgnoredDuringExecution：软策略,尝试满足,但是不保证.

---

举例：

apiVersion: v1
kind: Pod
metadata:
  name: with-node-affinity
spec:
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: topology.kubernetes.io/zone
            operator: In
            values:
            - antarctica-east1
            - antarctica-west1
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 1
        preference:
          matchExpressions:
          - key: another-node-label-key
            operator: In
            values:
            - another-node-label-value
  containers:
  - name: with-node-affinity
    image: k8s.gcr.io/pause:2.0

注意：

• 如果同时使用了nodeSelector和nodeAffinity,那么k8s在调度PodBoth conditions must be met at the same time.
• 如果在使用nodeAffinitywhen specifying multiplenodeSelectorTerms,Then the node only needs to satisfy one of themnodeSelectorTerms,can be dispatchedPod.
• 如果在一个nodeSelectorTermsMore than one is specified belowmatchExpressions,Then the node must satisfy allmatchExpressions,to be dispatchedPod.

污点与污点容忍

基于节点标签分配是站在Pod的角度上,通过在Pod上添加属性,来确定PodWhether to schedule to the specified node.相反地,我们也可以在NodeAdd a taint attribute to the node（Taints）,来避免Pod被分配到不合适的节点上.

• taints：避免Pod调度到特定节点上;
• tolerations：允许Pod调度到有Taints的节点上.

---

举例：

1. 给节点添加污点

#kubectl taint node <node名称> key=value:[effect]
kubectl taint node k8s-node2 gpu=yes:NoSchedule

其中,effect可以取值为：

• NoSchedule：一定不能被调度;
• PreferNoSchedule：尽量不要被调度,非必须;
• NoExecute：不仅不会调度,It will also evict existing ones on the nodePod.

2. 验证

kubectl describe node k8s-node2 | grep Taint

3. 如果希望PodCan be assigned to tainted nodes,要在PodAdd taint tolerance to configuration.

apiVersion: v1
kind: Pod
metadata:
  name: nginx
  labels:
    env: test
spec:
  containers:
  - name: nginx
    image: nginx
    imagePullPolicy: IfNotPresent
  tolerations:
  - key: "gpu"
    operator: "Equal"
	value: "yes"
    effect: "NoSchedule"

删除污点：

#kubectl taint node <node名称> key:[effect]-
kubectl taint node k8s-node2 gpu:NoSchedule-

注意：
Taints和Tolerations匹配的原则是key相同、effect相同,并且满足：

• 运算符是Exists（即没有value）;
• 运算符是Equal,且value相等.

两个特例：

• 一个空的key和运算符Exists,会匹配所有的key、value和effect,Means tolerating all blemishes;
• 一个key和一个空的effect匹配此key的所有effect.

References
【1】https://kubernetes.io/docs/concepts/scheduling-eviction/assign-pod-node/
【2】https://kubernetes.io/docs/concepts/scheduling-eviction/taint-and-toleration/