当前位置： 首页 > news >正文

k8s-pod调度

news 2025/10/3 7:34:55

k8s-pod调度

  • 一、调度方式与流程
  • 二、定向调度
    • nodeName
    • nodeSelector
  • 三、亲和性调度
    • nodeAffinity
    • podAffinity
    • podAntiAffinity
  • 四、污点和容忍
    • Taints
    • Toleration
    • 创建一个daemonset控制器，启动一个pod（nginx），在master节点也可以运行

官方文档：https://kubernetes.io/zh-cn/docs/concepts/scheduling-eviction/kube-scheduler/

一、调度方式与流程

  • 自动调度：运行在哪个节点上完全由Scheduler经过一系列的算法计算得出
  • 定向调度：NodeName、NodeSelector
  • 亲和性调度：NodeAffinity、PodAffinity、PodAntiAffinity
  • 污点（容忍）调度：Taints、Toleration

调度的核心流程

1.筛选
从集群所有节点中排除不满足 Pod 运行条件的节点，得到候选节点列表

  • 节点资源是否充足（CPU、内存等满足 resources.requests）
  • 节点是否有污点与 Pod 的容忍度匹配
  • 节点选择器或节点亲和性规则是否匹配
  • 端口是否冲突、主机名是否匹配等

2.打分

  • 对候选节点按优先级排序，选择得分最高的节点。
    默认打分策略包括：
  • 资源利用率
  • 节点亲和性权重
  • Pod 亲和性 / 反亲和性
  • 节点拓扑分布（如跨可用区均衡部署）

二、定向调度

调度是强制的，这就意味着即使要调度的目标Node不存在，也会向上面进行调度，只不过pod运行失败而已

给节点加标签 kubectl label nodes
查看标签 kubectl get node --show-labels

nodeName

1.创建一个命名空间

[root@master sch]# kubectl create namespace sc

2.编写pod的yaml文件


[root@master sch]# cat pod.yaml 
apiVersion: v1
kind: Namespace
metadata:name: sc
---
apiVersion: v1
kind: Pod
metadata: name: huangnamespace: sclabels:stu: good
spec:nodeName: k8s-3  # 指定调度到k8s-3节点上containers: - name: huang-nginximage: nginxports:- containerPort: 80name: http-port

3.应用

[root@k8s-1 pod]# kubectl apply -f pod.yaml 
pod/huang created
[root@k8s-1 pod]# kubectl get pod -n sc 
NAME    READY   STATUS    RESTARTS   AGE
huang   1/1     Running   0          13s
[root@k8s-1 pod]# kubectl get pod -n sc -o wide
NAME    READY   STATUS    RESTARTS   AGE     IP             NODE    NOMINATED NODE   READINESS GATES
huang   1/1     Running   0          18s   10.224.13.82   k8s-3   <none>           <none>

nodeSelector

[root@k8s-1 pod]# kubectl label nodes k8s-2 nodeenv=pro
node/k8s-2 labeled
[root@k8s-1 pod]# kubectl label nodes k8s-3 nodeenv=test
node/k8s-3 labeled
[root@k8s-1 pod]# kubectl get node --show-labels
NAME    STATUS   ROLES                  AGE   VERSION    LABELS
k8s-1   Ready    control-plane,master   9d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-1,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node-role.kubernetes.io/master=,node.kubernetes.io/exclude-from-external-load-balancers=
k8s-2   Ready    worker                 8d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-2,kubernetes.io/os=linux,node-role.kubernetes.io/worker=worker,nodeenv=pro
k8s-3   Ready    worker                 8d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-3,kubernetes.io/os=linux,node-role.kubernetes.io/worker=worker,nodeenv=test

创建一个带有节点选择器的 Pod

[root@k8s-1 pod]# cat pod-nodeselector.yaml 
apiVersion: v1
kind: Pod
metadata:name: pod-nodeselectornamespace: sc
spec:nodeSelector: nodeenv: procontainers:- name: s-nginximage: nginximagePullPolicy: IfNotPresentports:- containerPort: 80name: http-port

验证调度

[root@k8s-1 pod]# kubectl apply -f pod-nodeselector.yaml 
pod/pod-nodeselector created
[root@k8s-1 pod]# kubectl get pods -n sc pod-nodeselector -o wide
NAME               READY   STATUS    RESTARTS   AGE   IP               NODE    NOMINATED NODE   READINESS GATES
pod-nodeselector   1/1     Running   0          45s   10.224.200.210   k8s-2   <none>           <none>[root@k8s-1 pod]# kubectl describe pod -n sc pod-nodeselector
Events:Type    Reason     Age   From               Message----    ------     ----  ----               -------Normal  Scheduled  71s   default-scheduler  Successfully assigned sc/pod-nodeselector to k8s-2Normal  Pulled     66s   kubelet            Container image "nginx" already present on machineNormal  Created    66s   kubelet            Created container s-nginxNormal  Started    66s   kubelet            Started container s-nginx

三、亲和性调度

nodeAffinity

以node为目标，解决pod可以调度到哪些node的问题

如果两个应用频繁交互，那就有必要利用亲和性让两个应用的尽可能的靠近，这样可以减少因网络通信而带来的性能损耗

1.给node节点打标签

[root@k8s-1 pod]# kubectl label nodes k8s-3  disktype=ssd
node/k8s-3 labeled
## 删除标签，标签后接-
[root@k8s-1 pod]# kubectl label nodes k8s-3  disktype-
node/k8s-3 unlabeled[root@k8s-1 pod]# kubectl label nodes k8s-3  disktype=ssd
node/k8s-3 labeled
[root@k8s-1 pod]#  kubectl get nodes --show-labels
NAME    STATUS   ROLES                  AGE   VERSION    LABELS
k8s-1   Ready    control-plane,master   9d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-1,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node-role.kubernetes.io/master=,node.kubernetes.io/exclude-from-external-load-balancers=
k8s-2   Ready    worker                 9d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-2,kubernetes.io/os=linux,node-role.kubernetes.io/worker=worker,nodeenv=pro
k8s-3   Ready    worker                 9d    v1.23.17   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,disktype=ssd,kubernetes.io/arch=amd64,kubernetes.io/hostname=k8s-3,kubernetes.io/os=linux,node-
role.kubernetes.io/worker=worker,nodeenv=test

2.启动pod去选择打了标签的node

[root@k8s-1 pod]# cat pod-nodeaffinity.yaml 
apiVersion: v1
kind: Pod
metadata:name: l-nginx
spec:affinity:nodeAffinity:requiredDuringSchedulingIgnoredDuringExecution:nodeSelectorTerms:- matchExpressions:- key: disktypeoperator: Invalues:- ssd            containers:- name: nginximage: nginximagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f pod-nodeaffinity.yaml 
pod/l-nginx created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS    RESTARTS          AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Running   0                 113s    10.224.13.106    k8s-3   <none>           <none>

podAffinity

以pod为目标，解决pod可以和哪些已存在的pod部署在同一个拓扑域中的问题

创建第一个pod

[root@k8s-1 pod]# cat lyb.yaml 
apiVersion: v1
kind: Pod
metadata:name: lyb-nginxlabels: workcity: guangzhouspec:containers:- name: nginximage: nginximagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f lyb.yaml 
pod/lyb-nginx created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS             RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Running            0               27m     10.224.13.106    k8s-3   <none>           <none>
lyb-nginx                     1/1     Running            0               35s     10.224.200.201   k8s-2   <none>           <none>

创建第2个pod xa-nginx,使用pod亲和性，调度到lyb-nginx所在node节点

[root@k8s-1 pod]# cat xa.yaml 
apiVersion: v1
kind: Pod
metadata:name: xa-nginxlabels: workcity: guangzhou
spec:affinity:podAffinity:requiredDuringSchedulingIgnoredDuringExecution:- labelSelector:matchExpressions:- key: workcityoperator: Invalues: ["shenzhen","changsha","guangzhou"]topologyKey: kubernetes.io/hostnamecontainers:- name: nginximage: nginximagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f xa.yaml 
pod/xa-nginx created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS    RESTARTS          AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Running            0               27m     10.224.13.106    k8s-3   <none>           <none>
lyb-nginx                     1/1     Running   0                 5m16s   10.224.200.201   k8s-2   <none>           <none>
xa-nginx                      1/1     Running   0                 8s      10.224.200.200   k8s-2   <none>           <none>

lyb-nginx Pod 的标签是 workcity: guangzhou，符合 labelSelector 规则，因此 xa-nginx 被调度到了 lyb-nginx 所在的 k8s-2 节点

podAntiAffinity

以pod为目标，解决pod不能和哪些已存在pod部署在同一个拓扑域
中的问题

当应用的采用多副本部署时，有必要采用反亲和性让各个应用实例打散分布在各个
node上，这样可以提高服务的高可用性

[root@k8s-1 pod]# cat c.yaml 
apiVersion: v1
kind: Pod
metadata:name: c-nginxlabels: workcity: shanghai
spec:affinity:podAntiAffinity:requiredDuringSchedulingIgnoredDuringExecution:- labelSelector:matchExpressions:- key: workcityoperator: Invalues: ["shanghai","shenzhen","guangzhou"]topologyKey: kubernetes.io/hostnamecontainers:- name: nginximage: nginximagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f c.yaml 
pod/c-nginx created
# 验证
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS             RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Running            0               27m     10.224.13.106    k8s-3   <none>           <none>
c-nginx                       1/1     Running            0               6s      10.224.13.84     k8s-3   <none>           <none>
lyb-nginx                     1/1     Running            0               15m     10.224.200.201   k8s-2   <none>           <none>
xa-nginx                      1/1     Running            0               10m     10.224.200.200   k8s-2   <none>           <none>

四、污点和容忍

污点的格式为： key=value:effect , key和value是污点的标签，effect描述污点的作用，支持如下三个选项：

  • PreferNoSchedule：kubernetes将尽量避免把Pod调度到具有该污点的Node上，除非没有其他节点可调度
  • NoSchedule：kubernetes将不会把Pod调度到具有该污点的Node上，但不会影响当前Node上已存在的Pod
  • NoExecute：kubernetes将不会把Pod调度到具有该污点的Node上，同时也会将Node上已存在的Pod驱离

Taints

污点理解为一个标签，打了污点标签的节点，调度器在调度pod的时候，尽量不会调度到有污点的节点上，使节点能够排斥一类特定的 Pod

查看污点

[root@k8s-1 pod]# kubectl describe node k8s-1|grep Taint
Taints: node-role.kubernetes.io/master:NoSchedule

定义一个污点

[root@k8s-1 pod]# kubectl taint nodes k8s-3 city=beijing:NoSchedule
node/k8s-3 tainted
[root@k8s-1 pod]# kubectl describe node k8s-3|grep Taint
Taints: city=beijing:NoSchedule

[root@k8s-1 pod]# cat pod-4.yaml 
apiVersion: v1
kind: Pod
metadata:name: pod-taint-test
spec:containers:- name: nginximage: nginx:latestimagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f pod-4.yaml 
pod/pod-taint-test created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS    RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Running   0               27m     10.224.13.106    k8s-3   <none>           <none>
c-nginx                       1/1     Running   0               13m     10.224.13.107    k8s-3   <none>           <none>
lyb-nginx                     1/1     Running   0               29m     10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                1/1     Running   0               11s     10.224.200.199   k8s-2   <none>           <none>
xa-nginx                      1/1     Running   0               24m     10.224.200.200   k8s-2   <none>           <none>

删除污点

[root@k8s-1 pod]# kubectl taint nodes k8s-3 city=beijing:NoSchedule-
node/k8s-3 untainted

NoExecute使用

[root@k8s-1 pod]# kubectl taint nodes k8s-3 city=fudao:NoExecute
node/k8s-3 tainted
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS              RESTARTS         AGE     IP               NODE    NOMINATED NODE   READINESS GATES
l-nginx                       1/1     Terminating         0                63m     10.224.13.106    k8s-3   <none>           <none>
lyb-nginx                     1/1     Running             0                36m     10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                1/1     Running             0                7m14s   10.224.200.199   k8s-2   <none>           <none>
xa-nginx                      1/1     Running             0                31m     10.224.200.200   k8s-2   <none>           <none>[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS        RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
lyb-nginx                     1/1     Running       0               36m     10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                1/1     Running       0               7m34s   10.224.200.199   k8s-2   <none>           <none>
xa-nginx                      1/1     Running       0               31m     10.224.200.200   k8s-2   <none>           <none>

[root@k8s-1 pod]# cat pod-4.yaml 
apiVersion: v1
kind: Pod
metadata:name: pod-taint-test-2
spec:containers:- name: nginximage: nginx:latestimagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f pod-4.yaml 
pod/pod-taint-test-2 created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                          READY   STATUS             RESTARTS         AGE     IP               NODE    NOMINATED NODE   READINESS GATES
lyb-nginx                     1/1     Running            0                39m     10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                1/1     Running            0                10m     10.224.200.199   k8s-2   <none>           <none>
pod-taint-test-2              1/1     Running            0                3s      10.224.200.197   k8s-2   <none>           <none>
xa-nginx                      1/1     Running            0                34m     10.224.200.200   k8s-2   <none>           <none>

Toleration

容忍度允许调度器将pod调度到带有对应污点的节点上

[root@k8s-1 pod]# cat pod-5.yaml 
apiVersion: v1
kind: Pod
metadata:name: pod-taint-test-tolerations-2
spec:containers:- name: nginximage: nginx:latestimagePullPolicy: IfNotPresenttolerations: - key: "city"operator: "Equal" value: "beijing"effect: "NoSchedule"
[root@k8s-1 pod]# kubectl apply -f pod-5.yaml 
pod/pod-taint-test-tolerations-2 created
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                           READY   STATUS    RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
lyb-nginx                      1/1     Running   0               44m     10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                 1/1     Running   0               15m     10.224.200.199   k8s-2   <none>           <none>
pod-taint-test-2               1/1     Running   0               4m40s   10.224.200.197   k8s-2   <none>           <none>
pod-taint-test-tolerations-2   1/1     Running   0               8s      10.224.13.109    k8s-3   <none>           <none>
xa-nginx                       1/1     Running   0               39m     10.224.200.200   k8s-2   <none>           <none>

禁用节点调度
节点状态从 Ready 变为 Ready,SchedulingDisabled（就绪但禁止调度）

[root@k8s-1 pod]# kubectl cordon k8s-3
node/k8s-3 cordoned
[root@k8s-1 pod]# kubectl get node
NAME    STATUS                     ROLES                  AGE   VERSION
k8s-1   Ready                      control-plane,master   9d    v1.23.17
k8s-2   Ready                      worker                 9d    v1.23.17
k8s-3   Ready,SchedulingDisabled   worker                 9d    v1.23.17
# 恢复
[root@k8s-1 pod]# kubectl uncordon k8s-3
node/k8s-3 uncordoned

驱逐节点所有 Pod
kubectl drain k8s-3 --ignore-daemonsets

忽略 DaemonSet 管理的 Pod（这类 Pod 通常需要在每个节点运行，无法被驱逐）

创建一个daemonset控制器，启动一个pod（nginx），在master节点也可以运行

1.查看master节点有哪些污点，然后去给nginx pod配置容忍

[root@k8s-1 pod]# kubectl describe node k8s-1|grep Taint
Taints: node-role.kubernetes.io/master:NoSchedule

2.使用daemonset控制器去启动nginx pod

[root@k8s-1 pod]# cat daemonset-nginx.yaml 
apiVersion: apps/v1
kind: DaemonSet
metadata:name: daemonset-nginx
spec:selector:matchLabels:app: dm-nginxtemplate:metadata:labels:app: dm-nginxspec:tolerations:  - key: node-role.kubernetes.io/master	# # 匹配 master 节点的污点 keyoperator: Exists	# 只要污点存在就容忍effect: NoSchedule  # 匹配污点的效果- key: cityoperator: Existseffect: NoSchedulecontainers:- name: dm-nginximage: nginx:latestimagePullPolicy: IfNotPresent
[root@k8s-1 pod]# kubectl apply -f daemonset-nginx.yaml 
daemonset.apps/daemonset-nginx configured
[root@k8s-1 pod]# kubectl get pod -o wide
NAME                           READY   STATUS    RESTARTS        AGE     IP               NODE    NOMINATED NODE   READINESS GATES
daemonset-nginx-dsmrn          1/1     Running   0               6s      10.224.231.215   k8s-1   <none>           <none>
daemonset-nginx-nft4q          1/1     Running   0               10s     10.224.200.213   k8s-2   <none>           <none>
daemonset-nginx-z4ljp          1/1     Running   0               14s     10.224.13.110    k8s-3   <none>           <none>
lyb-nginx                      1/1     Running   0               105m    10.224.200.201   k8s-2   <none>           <none>
pod-taint-test                 1/1     Running   0               76m     10.224.200.199   k8s-2   <none>           <none>
pod-taint-test-2               1/1     Running   0               65m     10.224.200.197   k8s-2   <none>           <none>
pod-taint-test-tolerations-2   1/1     Running   0               61m     10.224.13.109    k8s-3   <none>           <none>
xa-nginx                       1/1     Running   0               100m    10.224.200.200   k8s-2   <none>           <none>
http://www.dtcms.com/a/434947.html

相关文章：

  • 中国工商做年报网站石家庄seo网站优化公司
  • 帝国CMS作文网题目文学文章wap+pc自适应响应式模板PHP网站源码
  • 邢台移动网站建设费用wordpress获取用户id
  • 网站设计的技能上海哪家做网站关键词排名
  • 酒店网站建设报价详情wordpress %1$s
  • 网站做支付要多少钱做产品代理上哪个网站好
  • RHEL安装
  • 列出网站开发建设的步骤通过网站建设提高企业的
  • leetcode 77 组合
  • 推广系统建站高清视频素材下载网站
  • freertos教程
  • C语言计算n个矩阵乘法
  • 做导航网站淘宝客做销量的网站有哪些
  • 注册免费域名网站wordpress仿百度搜索主题
  • c++MFC 用boost.asio库写的串口功能，包含发送、异步接收、打开、重连、关闭功能
  • 万用表和示波器使用
  • OpenJDK 和 Oracle JDK 该如何选择
  • 【开题答辩全过程】以 python基于Hadoop的服装穿搭系统的设计与实现为例，包含答辩的问题和答案
  • Spark专题-第三部分：性能监控与实战优化（2）-分区优化
  • port-isolate 概念及题目
  • 24.grep 使用手册
  • 俄罗斯网站设计电商网站开发设计方案
  • 在duckdb 1.4中编译和使用postgresql协议插件duckdb-pgwire
  • 鸿蒙开发2--常用UI组件与@State状态管理入门
  • Spring Boot 配置属性松散绑定
  • 重庆网站产品推广浙江省建设厅证书查询
  • 【代码随想录day 32】 力扣 509.斐波那契数列
  • 工信部网站备案怎么登录青岛公司做网站的价格
  • C语言中的scanf函数（头文件、格式控制、取地址符号分析）
  • 洛谷——P2437 蜜蜂路线