首页 > 其他分享 >day01 容器化的几种架构方式-pod创建过程原理-Scheduler的多种调度策略总结 (1.1-1.3.2 )

day01 容器化的几种架构方式-pod创建过程原理-Scheduler的多种调度策略总结 (1.1-1.3.2 )

时间:2023-11-20 18:45:45浏览次数:39  
标签:node io kubernetes 1.3 day01 affinity Scheduler master pod

一、容器化的几种架构方式

1.、容器的由来及变迁史

Docker->Docker-compose ->Docer swarm ->Kubernetes

2、微服务容器化的几种解决方案

特性 Docker Swarm   Kubernetes
安装和集群配置 安装简单,集群不强大 但在很复杂,集群非常强大
图形用户界面 没有官方图形界面,依托第三方 自带功能完善的Dashboard
可伸缩性 高度可伸缩,比Kubernetes快 高可伸缩性和快速扩展
自动伸缩 不能自动缩放 可自动缩放
负载均衡 可自动平衡集群中容器之间的流量 基于iptables/ipvs(调度算法丰富)
规模 1000节点、50000容器 节点不超过5000、Pod总数不超过150000、容器总数不超过300000、每个节点的pod数量不超过100
核心功能 管理节点高可用、调度任务、服务发现、回滚、更新、通讯安全、容器HA、服务自愈、配置管理、健康检查、服务伸缩

资源调度、服务发现、服务编排、资源逻辑管理、服务自愈、安全配置管理、job任务支持、自动回滚、内部域名服务、健康检查、扩容伸缩、负载均衡、灰度升级、应用HA、容灾恢复

设计初衷 跨宿主机管理容器集群 支持分布式、服务化的应用架构

2.1、Docer Swarm架构图

Manager nodes

  • 维护一个集群的状态
  • 对service进行调度
  • 为Swarm提供外部可调用的API接口

Work nodes

  • 用来执行Task

2.2 Kubernetes架构图

  • Master Node:作为控制节点,对集群进行调度管理;Master Node 由API Server、Schheduler、Cluster State Store和Controller-Manger Server组成
  • Work Node:作为真正的工作节点,运行业务应用的容器;Work Node包含kubelet、kube proxy和Container Runtime

  • etcd:保存整个集群的状态
  • apiserver:提供资源操作的唯一入口,并提供认证、授权、访问控制、API注册和服务发现等机制
  • controller manager:负责维护集群的状态,比如:故障检测、自动扩展、滚动更新等
  • scheduler:负责资源的调度,安装预定的调度策略将Pod调度到相应的机器

  • kubelet:负责维护集群的生命周期,同时也负责Volume(csi)和网络(cni)的管理
  • Container runtime:负责镜像管理以及Pod和容器的真正运行(cri)
  • kube-proxy:负责为Service提供cluster内部的服务发现和负载均衡

3、部署方式选择

kubeadm

  • 简单优雅
  • 支持高可用
  • 升级方便
  • 官方支持

kubernetes二进制

  • 早期线上主流方式
  • 部署时间长
  • 稳定性高

Kubernetes-as-a-Service(公有云服务)

  • 阿里云:ACK
  • aws cloud: EKS

二、POD创建过程原理

 

  1. 用户通过kubelet或web端向apiserver发送创建Pod请求
  2. apiserver会做出认证/鉴权响应,然后检查信息并把数据存储到ETCD数据库,创建deployment资源并初始化
  3. controller-manager通过list-watch机制,检查发现新的deployment,将资源加到内部工作队列,然后检查发现资源没有关联的pod和replicaset,启用deployment controller创建replicaset资源,再通过replicaset controller创建pod
  4. controller-manager 创建完成后将deployment,replicaset,pod资源更新到etcd
  5. scheduler也通过list-watch机制,检测发现新的pod,并通过预选及优选策略算法,来计算pod最终可调度的node节点,并通过apiserver将数据更新至etcd
  6. kubelet 每隔20s向apiserver通过Nodename获取自身node上所需要的pod清单;通过与自己的内部缓存进行比较,如果有新的资源则触发钩子调用CNI接口给pod创建pod网络,调用CRI接口启动容器,调用CSI进行存储卷的挂载,然后启动pod容器
  7. kube-proxy为新创建的pod注册动态dns到CoreOS。给oid的service添加iptables/ipvs规则,用于服务发现和负载均衡
  8. controller通过control loop(控制循环)将当前pod状态与用户所期望的状态做对比,如果当前状态与用户期望状态不同,则controller会将pod修改为用户期望状态,实在不行回将此pod删除,然后重新创建pod

三、Scheduler的多种调度策略总结上

1、简洁

Kubernetes 的调度去Scheduler,主要任务是把pod按照预定的策略分配到集群的节点上。

  • 公平:如何保证每个节点能分配到资源?
  • 资源高效利用:如何能把集群所有资源最大化利用
  • 效率:能够尽快地对大批量的pod完成调度工作
  • 灵活:允许特殊场景的应用调度到特定的节点

Sheduler是作为单独的程序运行,启动之后会一直监听apiserver,获取Podspec.NodeName 为空的Pod,对给个Pod都会创建一个binding,表明该pod应该放到哪个节点上。

2、调度过程

调度分为以下几个部分:

  1. 首先过滤掉不满足条件的节点,此过程成为预选(predicate)
  2. 然后对通过的节点按照优先级排序,此过程为优选(priority)
  3. 最后从中选择优先级最高的节点。如果中间任何一个步骤有错误,直接返回错误

Predicate有一些列算法使用:

  • PodFitsResources:节点上剩余的资源是否大于pod请求的资源
  • PodFitsHost:如果pod指定了NodeName,检查节点名称是否和NodeName匹配
  • PodFitsHostPorts:节点上已使用的port是否和pod申请的port冲突
  • PodselectorMatches:过滤掉和Pod指定的label不匹配的节点
  • NoDiskConflict:已经mount的volume和pod指定的volume不冲突,触发他们都是只读

如果predicate过程中没有合适的节点,pod会一直在pending状态,不断重试调度,直到有节点满足条件。如果有多个节点满足条件,就继续执行priorities过程:按照优先级大小对节点排序。

3、Node标签概念

每个node节点默认会有很多标签,标签在日常工作中就类似我们的一个标识;看到标签就知道这台node的主要用处

3.1 为节点加标签

[root@master-1-230 ~]# kubectl  get nodes
NAME           STATUS   ROLES           AGE     VERSION
master-1-230   Ready    control-plane   7d10h   v1.27.6
node-1-231     Ready    <none>          7d10h   v1.27.6
node-1-232     Ready    <none>          7d10h   v1.27.6
node-1-233     Ready    <none>          6d1h    v1.27.6

[root@master-1-230 ~]# kubectl get nodes --show-labels
NAME           STATUS   ROLES           AGE     VERSION   LABELS
master-1-230   Ready    control-plane   7d10h   v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=master-1-230,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node.kubernetes.io/exclude-from-external-load-balancers=
node-1-231     Ready    <none>          7d10h   v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,ingress=true,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-231,kubernetes.io/os=linux
node-1-232     Ready    <none>          7d10h   v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,ingress=true,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-232,kubernetes.io/os=linux
node-1-233     Ready    <none>          6d1h    v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-233,kubernetes.io/os=linux

[root@master-1-230 ~]# kubectl  label nodes node-1-231  apptype=core
node/node-1-231 labeled

[root@master-1-230 ~]# kubectl  get nodes --show-labels
NAME           STATUS   ROLES           AGE     VERSION   LABELS
master-1-230   Ready    control-plane   7d10h   v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=master-1-230,kubernetes.io/os=linux,node-role.kubernetes.io/control-plane=,node.kubernetes.io/exclude-from-external-load-balancers=
node-1-231     Ready    <none>          7d10h   v1.27.6   apptype=core,beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,ingress=true,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-231,kubernetes.io/os=linux
node-1-232     Ready    <none>          7d10h   v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,ingress=true,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-232,kubernetes.io/os=linux
node-1-233     Ready    <none>          6d1h    v1.27.6   beta.kubernetes.io/arch=amd64,beta.kubernetes.io/os=linux,kubernetes.io/arch=amd64,kubernetes.io/hostname=node-1-233,kubernetes.io/os=linux

[root@master-1-230 ~]# kubectl  label nodes node-1-231  apptype-
node/node-1-231 unlabeled

4、调度策略

4.1 nodeSelect

 nodeSelect是节点选择约束的最简单推荐形式。可以将nodeSelector字段设置为希望的目标节点所具有的节点标签。Kubernetes 只会将Pod调度到指定的标签节点上。

4.1.1 没有设置nodeSelect策略

# cat pod-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: busybox
spec:
  replicas: 3
  selector:
    matchLabels:
      app: busybox
  template:
    metadata:
      labels:
        app: busybox
    spec:
      containers:
      - name: busybox
        image: busybox
        imagePullPolicy: IfNotPresent
        command:
        - /bin/sh
        - -c
        - sleep 3000
[root@master-1-230 1.3]# kubectl  apply -f pod-demo.yaml 
deployment.apps/busybox created
[root@master-1-230 ~]# kubectl  get pod -o wide
NAME                       READY   STATUS    RESTARTS       AGE    IP               NODE         NOMINATED NODE   READINESS GATES
busybox-7bb7774f46-qjn8v   1/1     Running   0              22s    10.244.29.4      node-1-232   <none>           <none>
busybox-7bb7774f46-vtc75   1/1     Running   0              22s    10.244.154.6     node-1-233   <none>           <none>
busybox-7bb7774f46-wv5dt   1/1     Running   0              22s    10.244.167.135   node-1-231   <none>           <none>
testdp-85858fd689-bgn5s    1/1     Running   2 (139m ago)   7d1h   10.244.167.131   node-1-231   <none>           <none>

发现busybox pod运行在3个的node上

删除pod资源

[root@master-1-230 1.3]# kubectl  delete -f pod-demo.yaml 
deployment.apps "busybox" deleted

4.1.2 设置nodeSelector再观察运行结果

# cat pod-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: busybox
spec:
  replicas: 3
  selector:
    matchLabels:
      app: busybox
  template:
    metadata:
      labels:
        app: busybox
    spec:
      nodeSelector:
          apptype: core       ## 选择节点为core的标签
      containers:
      - name: busybox
        image: busybox
        imagePullPolicy: IfNotPresent
        command:
        - /bin/sh
        - -c
        - sleep 3000
[root@master-1-230 1.3]# kubectl  apply -f pod-demo.yaml 
deployment.apps/busybox created
[root@master-1-230 1.3]# kubectl  get pod -o wide
NAME                       READY   STATUS    RESTARTS       AGE    IP               NODE         NOMINATED NODE   READINESS GATES
busybox-7668d5bbd7-75j56   1/1     Running   0              36s    10.244.167.136   node-1-231   <none>           <none>
busybox-7668d5bbd7-cm5hl   1/1     Running   0              36s    10.244.167.137   node-1-231   <none>           <none>
busybox-7668d5bbd7-gzftw   1/1     Running   0              36s    10.244.167.138   node-1-231   <none>           <none>

pod busybox 运行在指定标签节点(apptype: core)

4.2 亲和性和反亲和性

使用亲和性与反亲和性的一些好处:

  • 亲和性、反亲和性语言的表达能力强。nodeSelect 只能选择所有固定标签的节点

亲和性:

  • requireDuringSchedulingInoredDuringExecution:调度器只有在规则满足的时候才会执行调度。此功能类型nodeSelector,但语言表达能力更强
  • preferredDuringSchedulingIgnoredDuringExecution:调度器会尝试寻找满足对应规则的节点。如果找不到匹配的节点,调度器仍然会在其他节点调度该Pod

NodeAffinity(节点亲和性):

  • 如果pod标明调度策略规则是“软需求”,这样调度器在无法找到匹配节点时仍然调度该Pod
  • 如果pod标明调度策略规则是“硬要求”,如果node解放店不满足需求,pod不会被调度,而是一直处于pending状态

PodAffinity(亲和性)/PodAntiAffinity(互斥性):

  • Pod亲和性主要解决pod可以和哪些pod部署在同一个拓扑域中的问题
  • Pod互斥性主要解决pod不能和哪些pod部署在同一个拓扑域中的问题

Kubernetes提供的操作符有以下几种:

调度策略 匹配符标签 操作符 拓扑域支持 调度目标
nodeAffinity node节点

In,NotIn,Exists,DoesNotExist.

Gt,Lt

指定主机
podAffinity pod In,NotIn,Exists,DoesNotExist Pod与指定Pod同一拓扑域

podAnitAffienity

pod

In,NotIn,Exists,DoesNotExist Pod与指定Pod不在同一拓扑域

Kubernetes提供的操作符有以下几种:

  • In:label的值在某个列表
  • NotIn:label的值不在某个列表中
  • Gt:label的值大于某个值
  • Lt:label的值小于某个值
  • Exists:某个label的值小于某个值
  • DoesNotExist:某个label不存在

验证:

NodeAffinity:

[root@master-1-230 1.3]# cat pod-nodeaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 3
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
            nodeSelectorTerms:
            - matchExpressions:
              - key: kubernetes.io/hostname
                operator: In
                values:
                - node-1-233
          preferredDuringSchedulingIgnoredDuringExecution:  # 软策略
          - weight: 1
            preference:
              matchExpressions:
              - key: apptype
                operator: In
                values:
                - core
[root@master-1-230 1.3]# kubectl  apply -f pod-nodeaff-demo.yaml 
deployment.apps/affinity created
[root@master-1-230 1.3]# kubectl  get pod -o wide
NAME                        READY   STATUS    RESTARTS       AGE    IP               NODE         NOMINATED NODE   READINESS GATES
affinity-7f586f85d4-7wpqs   1/1     Running   0              29s    10.244.154.7     node-1-233   <none>           <none>

podAffinity:

[root@master-1-230 1.3]# cat pod-podaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 15
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        podAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
          - labelSelector:
              matchExpressions:
              - key: app
                operator: In
                values:
                - busybox
            topologyKey: kubernetes.io/hostname
[root@master-1-230 1.3]# kubectl  apply -f pod-podaff-demo.yaml 
deployment.apps/affinity configured
[root@master-1-230 1.3]# 
[root@master-1-230 1.3]# kubectl  get pod -o wide
NAME                        READY   STATUS    RESTARTS   AGE     IP               NODE         NOMINATED NODE   READINESS GATES
affinity-747849f989-g86rg   1/1     Running   0          51s     10.244.167.142   node-1-231   <none>           <none>
busybox-7668d5bbd7-4694l    1/1     Running   0          3m57s   10.244.167.139   node-1-231   <none>           <none>
busybox-7668d5bbd7-dztrb    1/1     Running   0          3m57s   10.244.167.141   node-1-231   <none>           <none>
busybox-7668d5bbd7-qpbmk    1/1     Running   0          3m57s   10.244.167.140   node-1-231   <none>           <none>
testdp-85858fd689-pdh2q     1/1     Running   0          3m      10.244.29.5      node-1-232   <none>           <none>

podAntiAffinity

[root@master-1-230 1.3]# cat pod-podantiaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 15
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
          - labelSelector:
              matchExpressions:
              - key: app
                operator: In
                values:
                - busybox
            topologyKey: kubernetes.io/hostname
[root@master-1-230 1.3]# kubectl  apply -f pod-podantiaff-demo.yaml 
deployment.apps/affinity configured
[root@master-1-230 1.3]# kubectl  get pod -o wide
NAME                       READY   STATUS    RESTARTS   AGE   IP               NODE         NOMINATED NODE   READINESS GATES
affinity-564f689dc-9j4cw   1/1     Running   0          65s   10.244.154.8     node-1-233   <none>           <none>
busybox-7668d5bbd7-4694l   1/1     Running   0          11m   10.244.167.139   node-1-231   <none>           <none>
busybox-7668d5bbd7-dztrb   1/1     Running   0          11m   10.244.167.141   node-1-231   <none>           <none>
busybox-7668d5bbd7-qpbmk   1/1     Running   0          11m   10.244.167.140   node-1-231   <none>           <none>


实验升级:

1、如果不满足“Node硬亲和”?

[root@master-1-230 not]# cat not_pod-nodeaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 3
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        nodeAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
            nodeSelectorTerms:
            - matchExpressions:
              - key: kubernetes.io/hostname
                operator: In
                values:
                - node-1-233111
[root@master-1-230 not]# kubectl  apply -f not_pod-nodeaff-demo.yaml 
deployment.apps/affinity created
[root@master-1-230 ~]# kubectl  get pod
NAME                        READY   STATUS    RESTARTS   AGE
affinity-59b9fdffd4-fcp29   0/1     Pending   0          7s
您在 /var/spool/mail/root 中有新邮件
[root@master-1-230 ~]# kubectl  get pod
NAME                        READY   STATUS    RESTARTS   AGE
affinity-59b9fdffd4-fcp29   0/1     Pending   0          26s
[root@master-1-230 ~]# kubectl  describe pod affinity-59b9fdffd4-fcp29
Name:             affinity-59b9fdffd4-fcp29
Namespace:        default
Priority:         0
Service Account:  default
Node:             <none>
Labels:           app=affinity
                  pod-template-hash=59b9fdffd4
Annotations:      <none>
Status:           Pending
IP:               
IPs:              <none>
Controlled By:    ReplicaSet/affinity-59b9fdffd4
Containers:
  nginx:
    Image:        nginx:1.7.9
    Port:         80/TCP
    Host Port:    0/TCP
    Environment:  <none>
    Mounts:
      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-7g49m (ro)
Conditions:
  Type           Status
  PodScheduled   False 
Volumes:
  kube-api-access-7g49m:
    Type:                    Projected (a volume that contains injected data from multiple sources)
    TokenExpirationSeconds:  3607
    ConfigMapName:           kube-root-ca.crt
    ConfigMapOptional:       <nil>
    DownwardAPI:             true
QoS Class:                   BestEffort
Node-Selectors:              <none>
Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
  Type     Reason            Age   From               Message
  ----     ------            ----  ----               -------
  Warning  FailedScheduling  57s   default-scheduler  0/4 nodes are available: 1 node(s) had untolerated taint {node-role.kubernetes.io/control-plane: }, 3 node(s) didn't match Pod's node affinity/selector. preemption: 0/4 nodes are available: 4 Preemption is not helpful for scheduling..

2、如果不满足“node软亲和”?

[root@master-1-230 not]# cat not1_pod-nodeaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 3
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        nodeAffinity:
          preferredDuringSchedulingIgnoredDuringExecution:  # 软策略
          - weight: 1
            preference:
              matchExpressions:
              - key: apptype
                operator: In
                values:
                - corenot
[root@master-1-230 not]# kubectl  apply -f not1_pod-nodeaff-demo.yaml 
deployment.apps/affinity created
[root@master-1-230 not]# kubectl  get pod -o wide
NAME                        READY   STATUS    RESTARTS   AGE   IP             NODE         NOMINATED NODE   READINESS GATES
affinity-7bc8469b66-4f49x   1/1     Running   0          19s   10.244.154.9   node-1-233   <none>           <none>
[root@master-1-230 not]# 

3、如果不满足“pod硬亲和”?

[root@master-1-230 not]# cat  not_pod-podaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 15
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        podAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
          - labelSelector:
              matchExpressions:
              - key: app
                operator: In
                values:
                - busyboxnot
            topologyKey: kubernetes.io/hostname
[root@master-1-230 not]# 
[root@master-1-230 not]# kubectl  get pod
No resources found in default namespace.
您在 /var/spool/mail/root 中有新邮件
[root@master-1-230 not]# kubectl  apply -f not_pod-podaff-demo.yaml 
deployment.apps/affinity created
[root@master-1-230 not]# kubectl  get pod
NAME                        READY   STATUS    RESTARTS   AGE
affinity-5c66b5c66b-jp8sv   0/1     Pending   0          14s
[root@master-1-230 not]# kubectl  describe pod affinity-5c66b5c66b-jp8sv 
Name:             affinity-5c66b5c66b-jp8sv
Namespace:        default
Priority:         0
Service Account:  default
Node:             <none>
Labels:           app=affinity
                  pod-template-hash=5c66b5c66b
Annotations:      <none>
Status:           Pending
IP:               
IPs:              <none>
Controlled By:    ReplicaSet/affinity-5c66b5c66b
Containers:
  nginx:
    Image:        nginx:1.7.9
    Port:         80/TCP
    Host Port:    0/TCP
    Environment:  <none>
    Mounts:
      /var/run/secrets/kubernetes.io/serviceaccount from kube-api-access-9tfsn (ro)
Conditions:
  Type           Status
  PodScheduled   False 
Volumes:
  kube-api-access-9tfsn:
    Type:                    Projected (a volume that contains injected data from multiple sources)
    TokenExpirationSeconds:  3607
    ConfigMapName:           kube-root-ca.crt
    ConfigMapOptional:       <nil>
    DownwardAPI:             true
QoS Class:                   BestEffort
Node-Selectors:              <none>
Tolerations:                 node.kubernetes.io/not-ready:NoExecute op=Exists for 300s
                             node.kubernetes.io/unreachable:NoExecute op=Exists for 300s
Events:
  Type     Reason            Age   From               Message
  ----     ------            ----  ----               -------
  Warning  FailedScheduling  39s   default-scheduler  0/4 nodes are available: 1 node(s) had untolerated taint {node-role.kubernetes.io/control-plane: }, 3 node(s) didn't match pod affinity rules. preemption: 0/4 nodes are available: 4 Preemption is not helpful for scheduling..

4、如果不满足“pod软亲和”?

[root@master-1-230 not]# cat  not1_pod-podantiaff-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: affinity
  labels:
    app: affinity
spec:
  replicas: 1
  revisionHistoryLimit: 15
  selector:
    matchLabels:
      app: affinity
  template:
    metadata:
      labels:
        app: affinity
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - containerPort: 80
          name: nginxweb
      affinity:
        podAntiAffinity:
          requiredDuringSchedulingIgnoredDuringExecution:  # 硬策略
          - labelSelector:
              matchExpressions:
              - key: app
                operator: In
                values:
                - busyboxnot1
            topologyKey: kubernetes.io/hostname
[root@master-1-230 not]# kubectl  apply -f not1_pod-podantiaff-demo.yaml 
deployment.apps/affinity created
[root@master-1-230 not]# kubectl  get pod
NAME                        READY   STATUS    RESTARTS   AGE
affinity-6cb4595d67-j48h8   1/1     Running   0          11s

四、Scheduler的多种调度策略总结下

1、污点(taints)与容忍(tolerations)

对于nodeAffinity 无论是硬策略还是软策略方式,都是调度pod到预期节点,而Taints恰好与之相反,如果一个节点标记为Taints,除非pod也被表示为可以容忍,否则该Taints节点不可以被调度pod。节点被设置了污点,正常pod都不会被调度到该节点。

场景:

  1. 很多企业有基于GPU的算力集群,GPU的节点是不能让普通pod调度,需要对GPU的节点打上污点,防止pod被调度。
  2. 用户希望把Master节点暴露给Kubernetes系统组件使用,kubeadm搭建的集群默认给master节点添加一个污点标记,属于pod都没有被调度到master节点。

污点组成:key=value:effect

每个污点都有一个key和value作为污点的标签,其中value可以为空,effect描述污点的作用。当前taint effect支持以下三个选项:

  • NoSchedule:表示k8s不会将Pod调度到具有该污点的Node上
  • PreferNoSchedule:k8s将尽量避免将Pod调度到具有该污点的Node上。
  • NoExecute:k8s将不会将Pod调度到具有该污点的Node上,同时会将Node上已经存在的Pod驱逐出去。
#设置污点
[root@master-1-230 1.3]# kubectl taint nodes node-1-231 devops=node-1-231:NoSchedule
node/node-1-231 tainted

#查看污点
[root@master-1-230 1.3]# kubectl  describe nodes |grep -P "Name:|Taints"
Name:               master-1-230
Taints:             node-role.kubernetes.io/control-plane:NoSchedule
Name:               node-1-231
Taints:             devops=node-1-231:NoSchedule
Name:               node-1-232
Taints:             <none>
Name:               node-1-233
Taints:             <none>

#删除污点
kubectl[root@master-1-230 1.3]# kubectl taint nodes node-1-231 devops=node-1-231:NoSchedule-
node/node-1-231 untainted

实验:

[root@master-1-230 1.3]# cat pod-taint-demo.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: taint
  labels:
    app: taint
spec:
  replicas: 5
  revisionHistoryLimit: 10
  selector:
    matchLabels:
      app: taint
  template:
    metadata:
      labels:
        app: taint
    spec:
      containers:
      - name: nginx
        image: nginx:1.7.9
        ports:
        - name: http
          containerPort: 80
      tolerations:
      - key: "devops"
        operator: "Exists"
        effect: "NoSchedule"
[root@master-1-230 1.3]# kubectl apply -f  pod-taint-demo.yaml 
deployment.apps/taint created
[root@master-1-230 1.3]# kubectl  describe nodes |grep -P "Name:|Taints"
Name:               master-1-230
Taints:             node-role.kubernetes.io/control-plane:NoSchedule
Name:               node-1-231
Taints:             devops=node-1-231:NoSchedule
Name:               node-1-232
Taints:             <none>
Name:               node-1-233
Taints:             <none>
您在 /var/spool/mail/root 中有新邮件
[root@master-1-230 1.3]# kubectl  get pod -o wide
NAME                     READY   STATUS    RESTARTS   AGE   IP               NODE         NOMINATED NODE   READINESS GATES
taint-6cf8784b77-68nnr   1/1     Running   0          73s   10.244.154.11    node-1-233   <none>           <none>
taint-6cf8784b77-7sjd8   1/1     Running   0          73s   10.244.154.12    node-1-233   <none>           <none>
taint-6cf8784b77-dnx62   1/1     Running   0          73s   10.244.167.144   node-1-231   <none>           <none>
taint-6cf8784b77-sxfts   1/1     Running   0          73s   10.244.167.143   node-1-231   <none>           <none>
taint-6cf8784b77-vcztj   1/1     Running   0          73s   10.244.29.6      node-1-232   <none>           <none>

注意:

对应tolerations属性的写法,其中 key、value、effect与Node的Taint设置需要保持一致。

  • 如果operator的值是Exists,则value属性可省略
  • 如果operator的值是Equal,则表示其key与value之间的关系是equal(等于)
  • 如果不指定operator属性,则默认值是Equal

两个特殊的值:

  1. 空的key如果再配合Exists就能匹配所有的key与value,也能容忍所有node的所有污点(Taints)
  2. 空的effect匹配所有的effect

4.4 指定pod运行在固定节点

Pod.spec.nodeName 将Pod直接调度到指定Node节点上,会跳过Scheduler的调度策略,该匹配规则是强制匹配。

[root@master-1-230 1.3]# cat 6nodename.yaml 
apiVersion: apps/v1
kind: Deployment
metadata:
  name: myweb
spec:
  selector:
    matchLabels:
      app: myweb
  replicas: 3
  template:
    metadata:
      labels:
        app: myweb
    spec:
      nodeName: node-1-231       # 指定固定的节点
      containers:
      - name: myweb
        image: nginx
        ports:
        - containerPort: 80
[root@master-1-230 1.3]# kubectl  apply -f 6nodename.yaml 
deployment.apps/myweb created
您在 /var/spool/mail/root 中有新邮件
[root@master-1-230 1.3]# kubectl  get pod 
NAME                     READY   STATUS    RESTARTS   AGE
myweb-6dbb8cffc7-5bnwz   1/1     Running   0          50s
myweb-6dbb8cffc7-lnv9l   1/1     Running   0          50s
myweb-6dbb8cffc7-wqknf   1/1     Running   0          50s
taint-6cf8784b77-68nnr   1/1     Running   0          7m41s
taint-6cf8784b77-7sjd8   1/1     Running   0          7m41s
taint-6cf8784b77-dnx62   1/1     Running   0          7m41s
taint-6cf8784b77-sxfts   1/1     Running   0          7m41s
taint-6cf8784b77-vcztj   1/1     Running   0          7m41s

 

五、总结

nodeName:调度到固定节点;场景:验证特定节点上的数据/故障复现等

nodeSelect:会将Pod调度到拥有所指定的每个标签的节点上;场景:集群中的节点涉及特殊部门/业务线/测试情况

podAffinity:Pod亲和性主要解决pod可能和哪些pod部署在同一个拓扑域的问题;场景:上下游应用存在大量数据交互

podAntiAffinity:Pod互斥性主要解决pod不能和哪些pod部署在同一个拓扑域中的问题;场景:多个计算类应用不适合调度到同一台节点

nodeAffinity:指定应用可以调度到指定的节点;场景:核心应用必须调度到核心的节点

nodeAntiAffinity:指定应用不可以调度到指定的节点;场景:核心节点不允许被普通的应用调度

taints:该Taints节点不可以被调度pod;场景:GPU集群/master节点等

toleratons:配置了tolerations的pod是可以被调度到指定的Taints节点

GPU pod配置nodeAffinity必须要调度到指定的节点,同时该节点打了Taints不允许被其他pod调度

 

标签:node,io,kubernetes,1.3,day01,affinity,Scheduler,master,pod
From: https://www.cnblogs.com/pythonlx/p/17827552.html

相关文章

  • 魔兽争霸3冰封王座1.31下载
     魔兽争霸3冰封王座1.31下载链接:https://pan.baidu.com/s/1-if6LjzZVSd_MERuy9ZHpw提取码:shgx......
  • 文心一言 VS 讯飞星火 VS chatgpt (136)-- 算法导论11.3 2题
    二、用go语言,假设将一个长度为r的字符串散列到m个槽中,并将其视为一个以128为基数的数,要求应用除法散列法。我们可以很容易地把数m表示为一个32位的机器字,但对长度为r的字符串,由于它被当做以128为基数的数来处理,就要占用若干个机器字。假设应用除法散列法来计算一个字符串......
  • .NET8.0 AOT 经验分享 FreeSql/FreeRedis/FreeScheduler 均已通过测试
    2023年11月15日,对.net的开发圈是一个重大的日子,.net8.0正式版发布。圈内已经预热了有半个月有余,性能不断超越,开发体验越来越完美,早在.net5.0的时候就各种吹风Aot编译,直到6.07.0使用仍然比较麻烦,我个人比较期待本次更新的aot体验。有的群友几个小时都等不了啦,半夜就开始更新预......
  • idea2020.1.3汉化包报错问题
    已解决:idea2020.1.3汉化包报错问题问题描述:插件市场提供的版本不对、不兼容,所以需要手动下载安装这里附上文件使用方法:在插件市场,点击右上角设置,选择下面图片展示的选项,然后选择刚才下载的文件,点击重启即可完成!最后成品......
  • Q7.4.1.3. 产品销售 题解
    原题链接连\(S\toA_i\),流量\(D_i\),费用\(P_i\),表示最多进货\(D_i\),成本为\(P_i\)。连\(A_i\toT\),流量\(U_i\),费用\(0\),表示卖出。连\(A_i\toA_{i+1}\),流量\(+\infty\),费用\(C_i\),表示把\(A_i\)的货物拖一天花费\(C_i\)。连\(A_{i+1}\toA_i\),流量\(+\inft......
  • 文心一言 VS 讯飞星火 VS chatgpt (135)-- 算法导论11.3 1题
    一、用go语言,假设我们希望查找一个长度为n的链表,其中每一个元素都包含一个关键字k并具有散列值h(k)。每一个关键字都是长字符串。那么在表中查找具有给定关键字的元素时,如何利用各元素的散列值呢?文心一言:在Go语言中,你可以通过使用一个哈希表来存储关键字的散列值,并在查找具有给......
  • 文心一言 VS 讯飞星火 VS chatgpt (135)-- 算法导论11.3 1题
    一、用go语言,假设我们希望查找一个长度为n的链表,其中每一个元素都包含一个关键字k并具有散列值h(k)。每一个关键字都是长字符串。那么在表中查找具有给定关键字的元素时,如何利用各元素的散列值呢?文心一言:在Go语言中,你可以通过使用一个哈希表来存储关键字的散列值,并在查找具有......
  • 若智错误1.3
    很不好的消息,银狼所有的桥都被封掉了她一款一款游戏的登录,但是她忘了游戏名都有什么只好写出一个KMP模式匹配的板子去匹配游戏的名称,并且查询游戏名称在封禁名单里出现了几次点击查看代码#include<bits/stdc++.h>#definelCq<<1#definerCq<<1|1#defineintlonglong#......
  • 11.3
    -今天完成了一个分级测试,主要包括不同用户登录的选课系统,主要功能如下:(1)能够在Tomcat服务器中正确部署,并通过浏览器查看;(2)网站页面整体风格统一;(3)首页(登录页)要求实现不同用户登录后,进入的功能页不相同。(4)教师功能页:有添加课程、修改个人信息、浏览选课学生信息三个模块。(5)学......
  • 11.3
    运行时异常都是RuntimeException类及其子类异常,如NullPointerException、IndexOutOfBoundsException等,这些异常是不检查异常,程序中可以选择捕获处理,也可以不处理。这些异常一般由程序逻辑错误引起,程序应该从逻辑角度尽可能避免这类异常的发生。......