1.1 拓扑域划分

在超大规模集群中，可通过区域、可用区、数据中心、机柜等维度划分拓扑域，通过节点标签实现精准划分，示例如下：

# 按区域划分拓扑域（北京、南京区域）
kubectl label node k8s-master01 k8s-node01 region=beijing
kubectl label node k8s-node02 region=nanjing

# 按可用区划分（北京区域下分海淀、朝阳可用区）
kubectl label node k8s-master01 zone=beijing-haidian
kubectl label node k8s-node01 zone=beijing-chaoyang

# 按机房划分（海淀可用区下分不同机房）
kubectl label node k8s-master01 engineroom=beijing-haidian-c1

1.2 K8s 亲和力实战

1.2.1 同一个应用必须部署在不同的宿主机

通过podAntiAffinity的requiredDuringSchedulingIgnoredDuringExecution，强制应用副本不处于同一宿主机（拓扑域为宿主机 hostname）。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: diff-nodes  # 部署名称
  labels: 
    app: diff-nodes  # 应用标签，用于关联亲和性规则
spec: 
  selector: 
    matchLabels: 
      app: diff-nodes  # 选择匹配标签的Pod
  replicas: 2  # 部署2个副本
  template: 
    metadata: 
      labels: 
        app: diff-nodes  # Pod标签，与亲和性规则匹配
    spec: 
      affinity: 
        podAntiAffinity:  # Pod反亲和性：避免同一应用副本在同一拓扑域
          requiredDuringSchedulingIgnoredDuringExecution:  # 调度时必须满足，运行时忽略
          - labelSelector:  # 选择匹配的Pod标签
              matchExpressions: 
              - key: app  # 标签键
                operator: In  # 匹配规则：包含指定值
                values: [diff-nodes]  # 匹配的标签值
            topologyKey: kubernetes.io/hostname  # 拓扑域键：按宿主机划分
      containers: 
      - name: diff-nodes  # 容器名称
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12  # 容器镜像
        imagePullPolicy: IfNotPresent  # 镜像拉取策略：本地有则不拉取

1.2.2 同一个应用尽量部署在不同的宿主机

通过podAntiAffinity的preferredDuringSchedulingIgnoredDuringExecution，优先让应用副本不处于同一宿主机，不强制要求。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: diff-nodes  # 部署名称
  labels: 
    app: diff-nodes  # 应用标签
spec: 
  selector: 
    matchLabels: 
      app: diff-nodes  # 匹配Pod标签
  replicas: 2  # 2个副本
  template: 
    metadata: 
      labels: 
        app: diff-nodes  # Pod标签
    spec: 
      affinity: 
        podAntiAffinity:  # Pod反亲和性
          preferredDuringSchedulingIgnoredDuringExecution:  # 调度时优先满足，运行时忽略
          - podAffinityTerm:  # 亲和性条件
              labelSelector: 
                matchExpressions: 
                - key: app 
                  operator: In 
                  values: [diff-nodes]
              topologyKey: kubernetes.io/hostname  # 按宿主机划分拓扑域
            weight: 100  # 优先级权重（0-100，数值越高优先级越高）
      containers: 
      - name: diff-nodes 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12 
        imagePullPolicy: IfNotPresent

1.2.3 同一个应用分布在不同的机房

通过Pod反亲和性，优先让应用副本分布在不同可用域（拓扑域为zone），提升服务高可用性。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: diff-zone  # 部署名称（区分不同可用域）
  labels: 
    app: diff-zone  # 应用标签
spec: 
  selector: 
    matchLabels: 
      app: diff-zone  # 匹配Pod标签
  replicas: 3  # 3个副本，分布在不同可用域
  template: 
    metadata: 
      labels: 
        app: diff-zone  # Pod标签
    spec: 
      affinity: 
        podAntiAffinity:  # Pod反亲和性
          preferredDuringSchedulingIgnoredDuringExecution: 
          - podAffinityTerm: 
              labelSelector: 
                matchExpressions: 
                - key: app 
                  operator: In 
                  values: [diff-zone]
              topologyKey: zone  # 拓扑域键：按可用区划分
            weight: 100  # 高优先级
      containers: 
      - name: diff-nodes 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12 
        imagePullPolicy: IfNotPresent

1.2.4 应用尽量和缓存服务部署在同一个可用域

通过Pod亲和性，优先让应用与缓存服务部署在同一宿主机（拓扑域为hostname），提升访问性能。

# 1. 先部署缓存服务（标签为app: cache）
kubectl create deploy cache --image=registry.cn-beijing.aliyuncs.com/dotbalo/redis

# 2. 部署应用，通过亲和性关联缓存服务
apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: my-app  # 应用部署名称
spec: 
  replicas: 2  # 2个应用副本
  selector: 
    matchLabels: 
      app: my-app  # 应用Pod标签
  template: 
    metadata: 
      labels: 
        app: my-app  # Pod标签
    spec: 
      affinity: 
        podAffinity:  # Pod亲和性：应用与缓存服务尽量同域
          preferredDuringSchedulingIgnoredDuringExecution: 
          - weight: 100  # 高优先级
            podAffinityTerm: 
              labelSelector: 
                matchExpressions: 
                - key: app 
                  operator: In 
                  values: [cache]  # 匹配缓存服务的标签
              topologyKey: kubernetes.io/hostname  # 按宿主机划分
      containers: 
      - name: my-app 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12 
        imagePullPolicy: IfNotPresent

1.2.5 计算服务必须部署至高性能机器

通过nodeAffinity的requiredDuringSchedulingIgnoredDuringExecution，强制计算服务部署在指定高性能节点（标签disktype为ssd或nvme）。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: compute  # 计算服务部署名称
spec: 
  replicas: 2  # 2个计算服务副本
  selector: 
    matchLabels: 
      app: compute  # 应用标签
  template: 
    metadata: 
      labels: 
        app: compute  # Pod标签
    spec: 
      affinity: 
        nodeAffinity:  # 节点亲和性：指定部署节点
          requiredDuringSchedulingIgnoredDuringExecution:  # 调度时必须满足
            nodeSelectorTerms: 
            - matchExpressions: 
              - key: disktype  # 节点标签键（标识高性能磁盘）
                operator: In  # 匹配规则：包含指定值
                values: [ssd, nvme]  # 高性能磁盘类型
      containers: 
      - name: compute 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12 
        imagePullPolicy: IfNotPresent

1.2.6 计算服务尽量部署至高性能机器

通过nodeAffinity的preferredDuringSchedulingIgnoredDuringExecution，优先将计算服务部署在高性能节点，可设置权重区分优先级。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: compute  # 计算服务部署名称
spec: 
  replicas: 2 
  selector: 
    matchLabels: 
      app: compute 
  template: 
    metadata: 
      labels: 
        app: compute 
    spec: 
      affinity: 
        nodeAffinity:  # 节点亲和性
          preferredDuringSchedulingIgnoredDuringExecution:  # 优先满足
          # 优先级100：优先部署在ssd节点
          - weight: 100 
            preference: 
              matchExpressions: 
              - key: disktype 
                operator: In 
                values: [ssd]
          # 优先级50：其次部署在nvme节点
          - weight: 50
            preference: 
              matchExpressions: 
              - key: disktype 
                operator: In 
                values: [nvme]

1.2.7 应用尽量不部署至低性能机器

通过nodeAffinity，优先让应用不部署在低性能节点（标签performance为low），使用NotIn匹配规则。

apiVersion: apps/v1 
kind: Deployment 
spec: 
  name: compute-intensive-app  # 计算密集型应用名称
  replicas: 3 
  selector: 
    matchLabels: 
      app: compute-intensive  # 应用标签
  template: 
    metadata: 
      labels: 
        app: compute-intensive  # Pod标签
    spec: 
      affinity: 
        nodeAffinity:  # 节点亲和性
          preferredDuringSchedulingIgnoredDuringExecution: 
          - weight: 100 
            preference: 
              matchExpressions: 
              - key: performance  # 节点性能标签
                operator: NotIn  # 匹配规则：不包含指定值
                values: [low]  # 低性能节点标签值
      containers: 
      - name: compute-intensive 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12 
        imagePullPolicy: IfNotPresent

1.2.8 应用均匀分布在不同的机房

通过topologySpreadConstraints（拓扑域约束），确保应用副本在不同拓扑域（如宿主机）均匀分布，控制副本偏差。

apiVersion: apps/v1 
kind: Deployment 
metadata: 
  name: example-deployment  # 部署名称
spec: 
  replicas: 3  # 3个副本，均匀分布
  selector: 
    matchLabels: 
      app: example  # 应用标签
  template: 
    metadata: 
      labels: 
        app: example  # Pod标签
    spec: 
      topologySpreadConstraints:  # 拓扑域约束（多个约束需全部满足）
      - maxSkew: 1  # 最大副本偏差：任意拓扑域间副本数相差不超过1
        whenUnsatisfiable: DoNotSchedule  # 不满足约束时，不调度新Pod
        topologyKey: kubernetes.io/hostname  # 拓扑域键：按宿主机划分
        labelSelector:  # 应用约束的Pod标签选择器
          matchLabels: 
            app: example
      containers: 
      - name: example 
        image: registry.cn-beijing.aliyuncs.com/dotbalo/nginx:1.15.12
# 补充说明：
# whenUnsatisfiable可选值：
# - DoNotSchedule：不满足约束时，拒绝调度
# - ScheduleAnyway：不满足约束时，仍允许调度