kubernetes-什么可以实现pod调度？

作者：雪庆华
原创作品，严禁转载！

pod调度是什么？为什么要pod调度？
基于nodeName实现pod调度
基于hostNetwork实现pod调度
基于hostPort实现pod调度
基于resources实现pod调度
基于nodeSelector实现pod调度
基于taints实现pod调度
基于cordon实现pod调度
基于drain实现pod调度
基于Affinity亲和性实现pod调度

1.pod调度是什么？为什么要pod调度？

Pod 调度是 Kubernetes 中实现资源管理、高可用性、负载均衡和故障恢复的关键机制。通过合理的调度策略，Kubernetes 可以确保 Pod 在合适的节点上运行，同时满足应用的资源需求和部署约束。调度器的存在使得 Kubernetes 能够高效地管理大规模容器化应用，是 Kubernetes 集群管理的核心功能之一。
大白话就是：将pod与合适的节点关联

2.基于nodeName实现pod调度

通过节点名称，实现pod调度

案例展示：

  1.编写资源清单
[root@master231 scheduler]# cat  01-scheduler-nodeName.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-nodename
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: v1
    spec:
      nodeName: worker233
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
[root@master231 scheduler]# 

  2.验证 
[root@master231 scheduler]# kubectl apply  -f 01-scheduler-nodeName.yaml 
deployment.apps/deploy-xiuxian-nodename created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                      READY   STATUS    RESTARTS   AGE   IP             NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-nodename-966844f7d-cttm9   1/1     Running   0          39s   10.100.2.249   worker233   <none>           <none>
deploy-xiuxian-nodename-966844f7d-jzzvz   1/1     Running   0          39s   10.100.2.247   worker233   <none>           <none>
deploy-xiuxian-nodename-966844f7d-knqjs   1/1     Running   0          39s   10.100.2.246   worker233   <none>           <none>
deploy-xiuxian-nodename-966844f7d-q5td4   1/1     Running   0          39s   10.100.2.250   worker233   <none>           <none>
deploy-xiuxian-nodename-966844f7d-qcmzn   1/1     Running   0          39s   10.100.2.248   worker233   <none>           <none>
[root@master231 scheduler]#

3.基于hostNetwork实现pod调度

通过打开hostNetwork=true，允许 Pod 使用宿主机的网络命名空间

案例展示：

  1.编写资源清单
[root@master231 scheduler]# cat 03-scheduler-hostNetwork.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-hostnetwork
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: v1
    spec:
      hostNetwork: true
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
        ports:
        - containerPort: 80
[root@master231 scheduler]# 
  
  2.验证
[root@master231 scheduler]# kubectl apply -f 03-scheduler-hostNetwork.yaml
deployment.apps/deploy-xiuxian-hostnetwork created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get deploy,rs,pods -o wide
NAME                                         READY   UP-TO-DATE   AVAILABLE   AGE   CONTAINERS   IMAGES                                                      SELECTOR
deployment.apps/deploy-xiuxian-hostnetwork   2/5     5            2           6s    c1           registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1   apps=xiuxian

NAME                                                   DESIRED   CURRENT   READY   AGE   CONTAINERS   IMAGES                                                      SELECTOR
replicaset.apps/deploy-xiuxian-hostnetwork-bf5c7c4f6   5         5         2       6s    c1           registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1   apps=xiuxian,pod-template-hash=bf5c7c4f6

NAME                                             READY   STATUS    RESTARTS   AGE   IP           NODE        NOMINATED NODE   READINESS GATES
pod/deploy-xiuxian-hostnetwork-bf5c7c4f6-g4bs8   0/1     Pending   0          6s    <none>       <none>      <none>           <none>
pod/deploy-xiuxian-hostnetwork-bf5c7c4f6-gd9c5   1/1     Running   0          6s    10.0.0.233   worker233   <none>           <none>
pod/deploy-xiuxian-hostnetwork-bf5c7c4f6-jgztg   1/1     Running   0          6s    10.0.0.232   worker232   <none>           <none>
pod/deploy-xiuxian-hostnetwork-bf5c7c4f6-pkns2   0/1     Pending   0          6s    <none>       <none>      <none>           <none>
pod/deploy-xiuxian-hostnetwork-bf5c7c4f6-s5qd4   0/1     Pending   0          6s    <none>       <none>      <none>           <none>
[root@master231 scheduler]#

4.基于hostPort实现pod调度

hostPort 允许 Pod 内的容器直接使用宿主机的端口，从而允许外部网络通过宿主机的 IP 地址访问容器内的服务。

案例展示：

  1.编写资源清单
[root@master231 scheduler]# cat 02-scheduler-ports.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-ports
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: v1
    spec:
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
        ports:
        - containerPort: 80
          hostPort: 81
[root@master231 scheduler]# 

  2.验证
[root@master231 scheduler]# kubectl apply -f 02-scheduler-ports.yaml
deployment.apps/deploy-xiuxian-ports created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get deploy,rs,pods -o wide
NAME                                   READY   UP-TO-DATE   AVAILABLE   AGE   CONTAINERS   IMAGES                                                      SELECTOR
deployment.apps/deploy-xiuxian-ports   2/5     5            2           41s   c1           registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1   apps=xiuxian

NAME                                            DESIRED   CURRENT   READY   AGE   CONTAINERS   IMAGES                                                      SELECTOR
replicaset.apps/deploy-xiuxian-ports-f7fc745d   5         5         2       41s   c1           registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1   apps=xiuxian,pod-template-hash=f7fc745d

NAME                                      READY   STATUS    RESTARTS   AGE   IP            NODE        NOMINATED NODE   READINESS GATES
pod/deploy-xiuxian-ports-f7fc745d-26jmf   1/1     Running   0          41s   10.100.2.6    worker233   <none>           <none>
pod/deploy-xiuxian-ports-f7fc745d-dfvq7   1/1     Running   0          41s   10.100.1.54   worker232   <none>           <none>
pod/deploy-xiuxian-ports-f7fc745d-xw2d5   0/1     Pending   0          41s   <none>        <none>      <none>           <none>
pod/deploy-xiuxian-ports-f7fc745d-z2mts   0/1     Pending   0          41s   <none>        <none>      <none>           <none>
pod/deploy-xiuxian-ports-f7fc745d-zww84   0/1     Pending   0          41s   <none>        <none>      <none>           <none>
[root@master231 scheduler]#  

发现调度到两个节点，其他的pod处于pending状态，因为每台节点同一个端口只能匹配一各pod，由此间接实现pod调度

5.基于resources实现pod调度

resources表示为容器配置资源限制，通过资源限制，间接调度pod
不满足资源的节点，则不被调度
可实现的字段：requests期望资源、limits限制上限

requests案例展示：

[root@master231 scheduler]# cat 04-scheduler-resources-requests.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-resources
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
    spec:
      containers:
      - name: c1
        image: jasonyin2020/huazai007-linux-tools:v0.1
        # 为容器分配一个标准输入，类似于docker run -i 
        stdin: true
        # 为容器配置资源限制
        resources:
          # 表示用户的期望资源，符合期望才会被调度
          requests:
            cpu: 0.5
            # memory: 1G
            memory: 10G
[root@master231 scheduler]#

limits案例展示：

  1.编写资源清单
[root@master231 scheduler]# cat 05-scheduler-resources-limits.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-resources-limits
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: v1
    spec:
      containers:
      - name: c1
        image: jasonyin2020/huazai007-linux-tools:v0.1
        stdin: true
        # 如果不配置资源限制，则默认使用Pod调度节点的所有资源
        resources:
          # 配置资源的上限
          limits:
            cpu: 0.5
            memory: 2G
          # 配置容器的期望资源，如果不配置该字段，则默认和limits相同。
          # requests <= limits
          requests:
            # 1 core = 1000m
            cpu: 200m
            memory: 1G
[root@master231 scheduler]# 

压力测试命令:
  stress  -m 5 --vm-bytes 200000000 --vm-keep --verbose
可以进入容器内进行压力测试，验证限制是否成功
若超出上限，则会强制中断压制力测试

6.基于nodeSelector实现pod调度

nodeSelector可以基于节点标签调度Pod到对应的worker节点
使用方式：先给节点打上标签，再使用nodeSelector匹配节点，实现pod调度

案例展示：

    1 环境准备
[root@master231 scheduler]# kubectl label nodes worker232 school=huazai007
node/worker232 labeled
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl label nodes worker233 school=laonanhai
node/worker233 labeled
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes -l school --show-labels
NAME        STATUS   ROLES    AGE    VERSION    LABELS
worker232   Ready    <none>   2d3h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker232,kubernetes.io/os=linux,school=huazai007
worker233   Ready    <none>   2d3h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker233,kubernetes.io/os=linux,school=laonanhai
[root@master231 scheduler]# 


    2 编写资源清单
[root@master231 scheduler]# cat 06-scheduler-nodeSelector.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-nodeselector
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        class: beibei
    spec:
      # 基于节点标签调度Pod
      nodeSelector:
        school: laonanhai
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
[root@master231 scheduler]# 

    3 验证
[root@master231 scheduler]# kubectl apply -f 06-scheduler-nodeSelector.yaml
deployment.apps/deploy-xiuxian-nodeselector created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                           READY   STATUS    RESTARTS   AGE   IP            NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-nodeselector-7f78d9d78b-2gfcj   1/1     Running   0          4s    10.100.2.35   worker233   <none>           <none>
deploy-xiuxian-nodeselector-7f78d9d78b-fb88d   1/1     Running   0          4s    10.100.2.36   worker233   <none>           <none>
deploy-xiuxian-nodeselector-7f78d9d78b-rn96d   1/1     Running   0          4s    10.100.2.38   worker233   <none>           <none>
deploy-xiuxian-nodeselector-7f78d9d78b-s4jvf   1/1     Running   0          4s    10.100.2.34   worker233   <none>           <none>
deploy-xiuxian-nodeselector-7f78d9d78b-sfv7h   1/1     Running   0          4s    10.100.2.37   worker233   <none>           <none>
[root@master231 scheduler]#

7.基于taints实现pod调度

taints中译为污点。起作用就是，给节点打上污点，影响pod的调度。
格式为: "key[=value]:effect"

其中effect有三种类型:
NoSchedule
	不接受新的Pod，已经调度到该节点的Pod并不驱逐。
PreferNoSchedule:
	尽可能调度到其他节点，当其他节点不满足时，再调度到当前节点。
NoExecute:
	不接受新的Pod调度，与此同时，还会驱逐已经调度到给节点的所有Pod，不推荐使用。

7.1 taints的基本使用

1.查看污点

[root@master231 scheduler]# kubectl describe nodes master231  | grep Taints
Taints:             node-role.kubernetes.io/master:NoSchedule
[root@master231 scheduler]# 
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes worker232  | grep Taints
Taints:             <none>
[root@master231 scheduler]# 
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes worker233  | grep Taints
Taints:             <none>
[root@master231 scheduler]# 
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes  | grep Taints
Taints:             node-role.kubernetes.io/master:NoSchedule
Taints:             <none>
Taints:             <none>

2.打污点

kubectl taint node worker233 class=linux:NoSchedule

kubectl describe nodes  | grep Taints
Taints:             node-role.kubernetes.io/master:NoSchedule
Taints:             <none>
Taints:             class=linux:NoSchedule
kubectl taint node --all address=shahe:PreferNoSchedule
node/master231 tainted
node/worker232 tainted
node/worker233 tainted

[root@master231 deployments]# kubectl describe nodes  | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
                    address=shahe:PreferNoSchedule
Unschedulable:      false
--
Taints:             address=shahe:PreferNoSchedule
Unschedulable:      false
Lease:
--
Taints:             class=linux:NoSchedule
                    address=shahe:PreferNoSchedule
Unschedulable:      false

3.修改污点

kubectl taint node worker233 class=Linux:NoExecute --overwrite 
#--overwrite:表示覆盖原有污点

4.删除污点

kubectl taint node worker233 class=linux:NoSchedule-
#在原有污点后面加"-"

7.2验证污点作用

  1.书写资源清单
[root@master231 scheduler]# cat 07-scheduler-Taints.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-taints
  labels:
    school: huazai007
spec:
  replicas: 10
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: beibei
    spec:
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
        resources:
          requests:
            cpu: 0.2
            memory: 500Mi
[root@master231 scheduler]# 
  2.查看验证
在上面步骤已经将worker233打上污点了，那么pod将不往233调度
[root@master231 /oldboyedu/manifests/scheduler]# kubectl  describe  nodes  | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
--
Taints:             class=linux:NoSchedule
Unschedulable:      false
Lease:
[root@master231 /oldboyedu/manifests/scheduler]# kubectl  apply  -f 07-scheduler-Taints.yaml 
deployment.apps/deploy-xiuxian-taints created
[root@master231 /oldboyedu/manifests/scheduler]# kubectl  get pod  -o wide
NAME                                     READY   STATUS    RESTARTS        AGE   IP            NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-taints-85569f9fd5-4skpl   1/1     Running   0               10s   10.100.1.54   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-6r87l   1/1     Running   0               10s   10.100.1.52   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-89jt2   1/1     Running   0               10s   10.100.1.53   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-99kcf   1/1     Running   0               10s   10.100.1.56   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-cr9p9   1/1     Running   0               10s   10.100.1.51   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-grvlp   1/1     Running   0               10s   10.100.1.57   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-j7dfx   1/1     Running   0               10s   10.100.1.55   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-k9tfg   0/1     Pending   0               10s   <none>        <none>      <none>           <none>
deploy-xiuxian-taints-85569f9fd5-prwm6   0/1     Pending   0               10s   <none>        <none>      <none>           <none>
deploy-xiuxian-taints-85569f9fd5-v2cv7   0/1     Pending   0               10s   <none>        <none>      <none>           <none>
[root@master231 /oldboyedu/manifests/scheduler]# 

可以发现就算worker232节点没有资源了，也不忘worker233调度

7.3tolerations容忍污点

基于容忍污点实现pod调度
值得注意的是，一个Pod想要调度到一个有污点的worker node，则该Pod需要容忍该节点的所有污点，否则调度失败。

案例展示：

2.环境准备 
[root@master231 deployments]# kubectl describe nodes | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             school=huazai007:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             school=tiantian:NoExecute
                    class=beibei:NoSchedule
Unschedulable:      false
[root@master231 deployments]# 

3.编写资源清单
[root@master231 scheduler]# cat 08-scheduler-tolerations.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-tolerations
  labels:
    school: huazai007
spec:
  replicas: 10
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
    spec:
      # 配置污点容忍
      tolerations:
        # 匹配污点的key
      - key: node-role.kubernetes.io/master
        # 匹配污点的effect
        effect: NoSchedule
        # 表示key和value之间的关系，有效值为:Exists,Equal（default）。
        operator: Equal
      - key: school
        value: tiantian
        effect: NoExecute
        operator: Equal
      - key: class
        value: beibei
        effect: NoSchedule
        operator: Equal
      - key: school
        value: huazai007
        effect: NoSchedule
      # 无视任何污点，生产环境不推荐使用，临时测试可以.
      #tolerations:
      #- operator: Exists
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
        resources:
          requests:
            cpu: 0.2
            memory: 500Mi
[root@master231 scheduler]# 

后续检查pod状态，可以发现，实现了调度到以容忍的节点，比如master231

8.基于cordon实现pod调度

标记节点不可调度，与此同时会给改节点打污点。

使用方法：

2.实战案例
    2.1 环境准备
[root@master231 scheduler]# kubectl get nodes
NAME        STATUS   ROLES                  AGE    VERSION
master231   Ready    control-plane,master   2d5h   v1.23.17
worker232   Ready    <none>                 2d5h   v1.23.17
worker233   Ready    <none>                 2d5h   v1.23.17
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
[root@master231 scheduler]# 


    2.2 标记worker232节点不可调度 
[root@master231 scheduler]# kubectl cordon worker232 
node/worker232 cordoned
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes
NAME        STATUS                     ROLES                  AGE    VERSION
master231   Ready                      control-plane,master   2d5h   v1.23.17
worker232   Ready,SchedulingDisabled   <none>                 2d5h   v1.23.17
worker233   Ready                      <none>                 2d5h   v1.23.17
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             node.kubernetes.io/unschedulable:NoSchedule
Unschedulable:      true
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
[root@master231 scheduler]# 

后续验证，与上面案例一致

uncordon消除节点的污点、不可调度：

[root@master231 scheduler]# kubectl uncordon worker232 
node/worker232 uncordoned
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes
NAME        STATUS   ROLES                  AGE    VERSION
master231   Ready    control-plane,master   2d5h   v1.23.17
worker232   Ready    <none>                 2d5h   v1.23.17
worker233   Ready    <none>                 2d5h   v1.23.17
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
[root@master231 scheduler]#

9.基于drain实现pod调度

drain可以驱逐已经调度到节点的Pod。底层会调用cordon标记节点不可调度。
同样使用uncordon取消节点不可调度

案例展示：

    1 环境准备 
[root@master231 scheduler]# kubectl apply -f 07-scheduler-Taints.yaml 
deployment.apps/deploy-xiuxian-taints created
[root@master231 scheduler]# 
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                     READY   STATUS    RESTARTS   AGE   IP            NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-taints-85569f9fd5-2rhnc   1/1     Running   0          13s   10.100.2.92   worker233   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-4wmb5   1/1     Running   0          13s   10.100.1.96   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-55wzx   1/1     Running   0          13s   10.100.2.94   worker233   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-d88mt   1/1     Running   0          13s   10.100.1.97   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-k8b8b   1/1     Running   0          13s   10.100.2.95   worker233   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-pmx56   1/1     Running   0          13s   10.100.2.91   worker233   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-s85ls   1/1     Running   0          13s   10.100.2.93   worker233   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-wbtr8   1/1     Running   0          13s   10.100.1.98   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-xgsk9   1/1     Running   0          13s   10.100.1.95   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-zx8vk   1/1     Running   0          13s   10.100.1.94   worker232   <none>           <none>
[root@master231 scheduler]# 


    2 开始驱逐
[root@master231 scheduler]# kubectl drain worker233 --ignore-daemonsets 
node/worker233 already cordoned
WARNING: ignoring DaemonSet-managed Pods: kube-flannel/kube-flannel-ds-57zww, kube-system/kube-proxy-pzjt7
evicting pod default/deploy-xiuxian-taints-85569f9fd5-s85ls
evicting pod default/deploy-xiuxian-taints-85569f9fd5-55wzx
evicting pod default/deploy-xiuxian-taints-85569f9fd5-2rhnc
evicting pod default/deploy-xiuxian-taints-85569f9fd5-pmx56
evicting pod default/deploy-xiuxian-taints-85569f9fd5-k8b8b
pod/deploy-xiuxian-taints-85569f9fd5-55wzx evicted
pod/deploy-xiuxian-taints-85569f9fd5-s85ls evicted
pod/deploy-xiuxian-taints-85569f9fd5-2rhnc evicted
pod/deploy-xiuxian-taints-85569f9fd5-k8b8b evicted
pod/deploy-xiuxian-taints-85569f9fd5-pmx56 evicted
node/worker233 drained
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                     READY   STATUS    RESTARTS   AGE     IP             NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-taints-85569f9fd5-4wmb5   1/1     Running   0          3m50s   10.100.1.96    worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-5kmgj   0/1     Pending   0          47s     <none>         <none>      <none>           <none>
deploy-xiuxian-taints-85569f9fd5-5t4t8   0/1     Pending   0          47s     <none>         <none>      <none>           <none>
deploy-xiuxian-taints-85569f9fd5-d88mt   1/1     Running   0          3m50s   10.100.1.97    worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-gl67l   1/1     Running   0          47s     10.100.1.100   worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-l5jwj   1/1     Running   0          47s     10.100.1.99    worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-wbtr8   1/1     Running   0          3m50s   10.100.1.98    worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-xgsk9   1/1     Running   0          3m50s   10.100.1.95    worker232   <none>           <none>
deploy-xiuxian-taints-85569f9fd5-zrqfp   0/1     Pending   0          47s     <none>         <none>      <none>           <none>
deploy-xiuxian-taints-85569f9fd5-zx8vk   1/1     Running   0          3m50s   10.100.1.94    worker232   <none>           <none>
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes
NAME        STATUS                     ROLES                  AGE    VERSION
master231   Ready                      control-plane,master   2d5h   v1.23.17
worker232   Ready                      <none>                 2d5h   v1.23.17
worker233   Ready,SchedulingDisabled   <none>                 2d5h   v1.23.17
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl describe nodes | grep Taints -A 2
Taints:             node-role.kubernetes.io/master:NoSchedule
Unschedulable:      false
Lease:
--
Taints:             <none>
Unschedulable:      false
Lease:
--
Taints:             node.kubernetes.io/unschedulable:NoSchedule
Unschedulable:      true
Lease:
[root@master231 scheduler]#

10.基于Affinity亲和性实现pod调度

10.1nodeAffinity节点亲和度

nodeAffinity表示Pod调度到期望节点。
功能和nodeSelector类似，但功能更加强大。
nodeSelector的key只能匹配一个velue
nodeAffintify但是可以匹配多个

案例展示：

  1.书写资源清单
[root@master231 scheduler]# cat 09-scheduler-nodeAffinity.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-nodeaffinity
  labels:
    school: huazai007
spec:
  replicas: 10
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: beibei
    spec:
      tolerations:
      - operator: Exists
      # 定义亲和性
      affinity:
        # 定义节点亲和性
        nodeAffinity:
          # 定义硬限制
          requiredDuringSchedulingIgnoredDuringExecution:
            # 定义节点选择器
            nodeSelectorTerms:
              # 基于表达式匹配节点
            - matchExpressions:
                # 定义节点标签的key
              - key: dc
                # 定义节点标签的value
                values:
                - beijing
                - shanghai
                # 定义key和values之间的关系，有效值为: In,NotIn, Exists, DoesNotExist. Gt, and Lt
                operator: In
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes -l dc=beijing
NAME        STATUS   ROLES    AGE   VERSION
worker232   Ready    <none>   15h   v1.23.17
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get nodes -l dc=shanghai
NAME        STATUS   ROLES                  AGE     VERSION
master231   Ready    control-plane,master   2d21h   v1.23.17
[root@master231 scheduler]# 

  2.查看验证
[root@master231 scheduler]# kubectl  apply -f 09-scheduler-nodeAffinity.yaml 
deployment.apps/deploy-xiuxian-nodeaffinity created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                          READY   STATUS    RESTARTS   AGE   IP            NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-nodeaffinity-b7f465544-72fqm   1/1     Running   0          8s    10.100.0.26   master231   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-7xq55   1/1     Running   0          8s    10.100.1.21   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-bcrkc   1/1     Running   0          8s    10.100.1.24   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-brd4w   1/1     Running   0          8s    10.100.0.29   master231   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-fglx4   1/1     Running   0          8s    10.100.1.20   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-g9jc5   1/1     Running   0          8s    10.100.1.19   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-gb7mq   1/1     Running   0          8s    10.100.0.27   master231   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-pjsz9   1/1     Running   0          8s    10.100.1.22   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-pkhcg   1/1     Running   0          8s    10.100.1.23   worker232   <none>           <none>
deploy-xiuxian-nodeaffinity-b7f465544-qlsxl   1/1     Running   0          8s    10.100.0.28   master231   <none>           <none>
[root@master231 scheduler]#

10.2podAffinity

podAffinity在调度Pod时可以基于拓扑域进行调度，当首个Pod调度到特定的拓扑域后，后续的所有Pod都往该拓扑域调度。

在 Kubernetes 中，拓扑域（Topology Domain） 是一个用于描述集群中资源分布的逻辑层次结构的概念。
可以理解为"一个区域"。

通过标签实现控制pod后续调度

案例展示：

   1 环境准备 
[root@master231 scheduler]# kubectl get nodes --show-labels -l "dc in (beijing,shanghai,shenzhen)" | grep dc
master231   Ready    control-plane,master   2d23h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=shanghai,kubernetes.io/arch=amd64,kubernetes.io/hostname=master231,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node-role.kubernetes.io/master=,node.kubernetes.io/exclude-from-external-load-balancers
worker232   Ready    <none>                 16h     v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=beijing,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker232,kubernetes.io/os=linux
worker233   Ready    <none>                 2d22h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=shenzhen,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker233,kubernetes.io/os=linux,school=laonanhai
[root@master231 scheduler]# 


   2 编写资源清单
[root@master231 scheduler]# cat 10-scheduler-podAffinity.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-podaffinity
  labels:
    school: huazai007
spec:
  replicas: 10
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: beibei
    spec:
      tolerations:
      - operator: Exists
      # 定义亲和性
      affinity:
        # 定义Pod亲和性
        podAffinity:
          # 定义硬限制
          requiredDuringSchedulingIgnoredDuringExecution:
            # 定义拓扑域(“机房”)
          - topologyKey: dc
            # 定义Pod标签选择器
            labelSelector:
              # 基于表达式匹配节点
              matchExpressions:
                # 定义节点标签的key
              - key: apps
                # 定义节点标签的value
                values:
                - xiuxian
                # 定义key和values之间的关系，有效值为: In,NotIn, Exists, DoesNotExist. Gt, and Lt
                operator: In
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
[root@master231 scheduler]# 


   3 测试验证
[root@master231 scheduler]# kubectl apply -f 10-scheduler-podAffinity.yaml
deployment.apps/deploy-xiuxian-podaffinity created
[root@master231 scheduler]# 
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                          READY   STATUS    RESTARTS   AGE   IP             NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-podaffinity-7c64bd95f7-5564f   1/1     Running   0          11s   10.100.2.124   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-gsg7x   1/1     Running   0          11s   10.100.2.127   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-pgc5f   1/1     Running   0          11s   10.100.2.126   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-pl466   1/1     Running   0          11s   10.100.2.125   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-qcp6m   1/1     Running   0          11s   10.100.2.129   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-rplw5   1/1     Running   0          11s   10.100.2.128   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-rzhmm   1/1     Running   0          11s   10.100.2.121   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-ssc46   1/1     Running   0          11s   10.100.2.123   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-tjsrn   1/1     Running   0          11s   10.100.2.122   worker233   <none>           <none>
deploy-xiuxian-podaffinity-7c64bd95f7-whrvg   1/1     Running   0          11s   10.100.2.120   worker233   <none>           <none>
[root@master231 scheduler]#

10.3podAntiAffinity

podAntiAffinity和podAffinity相反。

在调度Pod时可以基于拓扑域进行调度，当某个Pod调度到特定的拓扑域后，后续的Pod都不能往该拓扑域调度。

案例展示：

  1 环境准备 
[root@master231 scheduler]# kubectl get nodes --show-labels -l "dc in (beijing,shanghai,shenzhen)" | grep dc
master231   Ready    control-plane,master   2d23h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=shanghai,kubernetes.io/arch=amd64,kubernetes.io/hostname=master231,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node-role.kubernetes.io/master=,node.kubernetes.io/exclude-from-external-load-balancers=,school=laonanhai
worker232   Ready    <none>                 16h     v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=beijing,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker232,kubernetes.io/os=linux
worker233   Ready    <none>                 2d22h   v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,dc=shenzhen,kubernetes.io/arch=amd64,kubernetes.io/hostname=worker233,kubernetes.io/os=linux,school=laonanhai
[root@master231 scheduler]# 
    
  2 编写资源清单
[root@master231 scheduler]# cat 11-scheduler-podAntiAffinity.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deploy-xiuxian-podantiaffinity
  labels:
    school: huazai007
spec:
  replicas: 5
  selector:
    matchLabels:
      apps: xiuxian
  template:
    metadata:
      labels:
        apps: xiuxian
        address: shahe
        class: beibei
    spec:
      tolerations:
      - operator: Exists
      # 定义亲和性
      affinity:
        # 定义Pod反亲和性
        podAntiAffinity:
          # 定义硬限制
          requiredDuringSchedulingIgnoredDuringExecution:
            # 定义拓扑域(“机房”)
          - topologyKey: dc
            # 定义Pod标签选择器
            labelSelector:
              # 基于表达式匹配节点
              matchExpressions:
                # 定义节点标签的key
              - key: apps
                # 定义节点标签的value
                values:
                - xiuxian
                # 定义key和values之间的关系，有效值为: In,NotIn, Exists, DoesNotExist. Gt, and Lt
                operator: In
      containers:
      - name: c1
        image: registry.cn-hangzhou.aliyuncs.com/huazai-k8s/apps:v1
[root@master231 scheduler]# 


   3 测试验证
[root@master231 scheduler]# kubectl get pods -o wide
NAME                                              READY   STATUS    RESTARTS   AGE   IP             NODE        NOMINATED NODE   READINESS GATES
deploy-xiuxian-podantiaffinity-7f774fc697-mzm8r   0/1     Pending   0          43s   <none>         <none>      <none>           <none>
deploy-xiuxian-podantiaffinity-7f774fc697-nt8xs   1/1     Running   0          43s   10.100.2.130   worker233   <none>           <none>
deploy-xiuxian-podantiaffinity-7f774fc697-pbtwx   1/1     Running   0          43s   10.100.1.51    worker232   <none>           <none>
deploy-xiuxian-podantiaffinity-7f774fc697-sw9gd   1/1     Running   0          43s   10.100.0.30    master231   <none>           <none>
deploy-xiuxian-podantiaffinity-7f774fc697-xtmjk   0/1     Pending   0          43s   <none>         <none>      <none>           <none>
[root@master231 scheduler]#

posted @ 2025-02-15 22:14 悍匪屋阅读(3) 评论(0) 编辑收藏举报

刷新页面返回顶部

登录后才能查看或发表评论，立即登录或者逛逛博客园首页

相关博文：

· kubernetes-常见的控制器有哪些?

· kubernets-基本使用

· K8S-集群调度

· k8s 亲和性反亲和性-污点和容忍度

· K8S调度约束

阅读排行：
· 全程不用写代码，我用AI程序员写了一个飞机大战
· DeepSeek 开源周回顾「GitHub 热点速览」
· 记一次.NET内存居高不下排查解决与启示
· MongoDB 8.0这个新功能碉堡了，比商业数据库还牛
· .NET10 - 预览版1新功能体验（一）

公告

昵称：悍匪屋
园龄： 1个月
粉丝： 0
关注： 0

+加关注

2025年3月

日

一

二

三

四

五

六

随笔档案

2025年2月(4)

阅读排行榜

1. ubuntu-kubernetes集群搭建(16)

xqingh

kubernetes-什么可以实现pod调度？

目录

1.pod调度是什么？为什么要pod调度？

2.基于nodeName实现pod调度

3.基于hostNetwork实现pod调度

4.基于hostPort实现pod调度

5.基于resources实现pod调度

6.基于nodeSelector实现pod调度

7.基于taints实现pod调度

7.1 taints的基本使用

7.2验证污点作用

7.3tolerations容忍污点

8.基于cordon实现pod调度

9.基于drain实现pod调度

10.基于Affinity亲和性实现pod调度

10.1nodeAffinity节点亲和度

10.2podAffinity

10.3podAntiAffinity

公告

搜索

常用链接

我的标签

随笔档案

阅读排行榜