【CKA】Kubernetes Pod调度

2019年10月15日16:31:43 发表评论 191 views

介绍

Pod的调度在默认情况下是 Scheduler Controller 采用默认算法的全自动调度，在实际使用中并不满足我们的需求，因为并不能事先掌握pod被调度到哪个Node之上，所以kubernetes又提供了好几种方式让我们自已选择调度到什么Node中，比如有NodeSelector(定向调度)、NodeAffinity(Node亲和性)、PodAffinity(Pod亲和性)、Taints

NodeSelector调度算法比较简单，NodeSelector 只调度到某个拥有特定标签的Node上，如果没有满足条件的Node，那么此Pod将不会被运行，即使在集群中还有可用Node列表，这就限制了它的使用场景，现在基本上被NodeAffinity取代了，NodeAffinity在NodeSelector的基础之上的进行了扩展，使调度更加灵活，除了有一个必须要满足条件的Node之外，还要可设置优先条件，下面来看下NodeAffinity的使用方式：

当前节点：

当前Pod：

1、nodeName

通过主机名指定pod调度到指定节点

[root@k8s-master 02]# vim manual-schedule.yaml
apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-manual-schedule
  name: pod-manual-schedule
  namespace: default
spec:
  nodeName: k8s-node2
  containers:
  - name: my-pod
    image: nginx
    imagePullPolicy: IfNotPresent

[root@k8s-master 02]# vim manual-schedule.yaml

apiVersion: v1

kind: Pod

metadata:

labels:

run: pod-manual-schedule

namespace: default

spec:

nodeName: k8s-node2

containers:

- name: my-pod

image: nginx

imagePullPolicy: IfNotPresent

[root@k8s-master 02]# kubectl apply -f manual-schedule.yaml

1	[root@k8s-master 02]# kubectl apply -f manual-schedule.yaml

2、nodeSelector

Kubernetes中常用label来管理集群的资源，nodeSelector可通过标签实现pod调度到指定节点上。

举列：使用 nodeSelector 将 pod 调度到 k8s-node2 节点上

Step1：给k8s-node2打标签

kubectl label nodes k8s-node2 test=nginx #设置标签
kubectl get nodes --show-labels |grep test=nginx #查询标签是否设置成功

1 2	kubectl label nodes k8s-node2 test=nginx #设置标签 kubectl get nodes --show-labels \|grep test=nginx #查询标签是否设置成功

Step2：nodeSelector设置对应标签

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-selector-schedule
  name: pod-selector-schedule
  namespace: default
spec:
  nodeName: k8s-node2
  containers:
  - name: my-pod
    image: nginx
    imagePullPolicy: IfNotPresent
  nodeSelector:
    test: nginx

apiVersion: v1

kind: Pod

metadata:

labels:

run: pod-selector-schedule

namespace: default

spec:

nodeName: k8s-node2

containers:

- name: my-pod

image: nginx

imagePullPolicy: IfNotPresent

nodeSelector:

test: nginx

3、nodeAffinity

node 节点 Affinity ，从字面上很容易理解nodeAffinity就是节点亲和性，Anti-Affinity也就是反亲和性。节点亲和性就是控制pod是否调度到指定节点，相对nodeSelector来说更为灵活，可以实现一些简单的逻辑组合。

preferredDuringSchedulingIgnoredDuringExecution #软策略，尽量满足
requiredDuringSchedulingIgnoredDuringExecution  #硬策略，必须满足

1 2	preferredDuringSchedulingIgnoredDuringExecution #软策略，尽量满足 requiredDuringSchedulingIgnoredDuringExecution #硬策略，必须满足

场景1：必须部署到有test=nginx(k8s-node2)标签的节点

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: node-affinity
  name: node-affinity
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: node-affinity
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: test
            operator: In
            values:
            - nginx

apiVersion: v1

kind: Pod

metadata:

labels:

run: node-affinity

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

affinity:

nodeAffinity:

requiredDuringSchedulingIgnoredDuringExecution:

nodeSelectorTerms:

- matchExpressions:

- key: test

operator: In

values:

- nginx

场景2：最好部署到有test=nginx(k8s-node2)标签的节点

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: node-affinity
  name: node-affinity
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: node-affinity
  affinity:
    nodeAffinity:
      preferredDuringSchedulingIgnoredDuringExecution:
      - weight: 1
        preference:
          matchExpressions:
          - key: test
            operator: In
            values:
            - nginx

apiVersion: v1

kind: Pod

metadata:

labels:

run: node-affinity

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

affinity:

nodeAffinity:

preferredDuringSchedulingIgnoredDuringExecution:

- weight: 1

preference:

matchExpressions:

- key: test

operator: In

values:

- nginx

场景3：不能部署在k8s-node2节点

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: node-affinity
  name: node-affinity
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: node-affinity
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: kubernetes.io/hostname
            operator: NotIn
            values:
            - k8s-node2

apiVersion: v1

kind: Pod

metadata:

labels:

run: node-affinity

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

affinity:

nodeAffinity:

requiredDuringSchedulingIgnoredDuringExecution:

nodeSelectorTerms:

- matchExpressions:

- key: kubernetes.io/hostname

operator: NotIn

values:

- k8s-node2

Kubernetes中的operator提供了下面几种过滤条件:

- In：label 的值在某个列表中
- 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 存在
- DoesNotExist：某个 label 不存

- In：label 的值在某个列表中

- 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 存在

- DoesNotExist：某个 label 不存

3、podAffinity

nodeSelector 和 nodeAffinity 都是控制 pod 调度到节点的操作，在实际项目部署场景中，希望根据服务与服务之间的关系进行调度，也就是根据 pod 之间的关系进行调度，比如同一个项目的前后端，希望可以在同一个节点上，Kubernetes的podAffinity就可以实现这样的场景，podAffinity的调度策略和nodeAffinity类似也有：

requiredDuringSchedulingIgnoredDuringExecution
preferredDuringSchedulingIgnoredDuringExecution

1 2	requiredDuringSchedulingIgnoredDuringExecution preferredDuringSchedulingIgnoredDuringExecution

场景：希望新建的pod-affinity 与 node-affinity运行在同一个节点上

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-affinity
  name: pod-affinity
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-affinity
  affinity:
    podAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: run
            operator: In
            values:
            - "node-affinity"
        topologyKey: kubernetes.io/hostname

apiVersion: v1

kind: Pod

metadata:

labels:

run: pod-affinity

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

affinity:

podAffinity:

requiredDuringSchedulingIgnoredDuringExecution:

- labelSelector:

matchExpressions:

- key: run

operator: In

values:

- "node-affinity"

topologyKey: kubernetes.io/hostname

4、podAntiAffinity

Pod反亲和性

场景：希望新建的pod-affinity 与 node-affinity运行在不同节点上

apiVersion: v1
kind: Pod
metadata:
  labels:
    run: pod-anti-affinity
  name: pod-anti-affinity
spec:
  containers:
  - image: nginx
    imagePullPolicy: IfNotPresent
    name: pod-anti-affinity
  affinity:
    podAntiAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - labelSelector:
          matchExpressions:
          - key: run
            operator: In
            values:
            - "pod-affinity"
        topologyKey: kubernetes.io/hostname

apiVersion: v1

kind: Pod

metadata:

labels:

run: pod-anti-affinity

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

affinity:

podAntiAffinity:

requiredDuringSchedulingIgnoredDuringExecution:

- labelSelector:

matchExpressions:

- key: run

operator: In

values:

- "pod-affinity"

topologyKey: kubernetes.io/hostname

5、Taints & tolerations

Taints：污点，

tolerations：容忍

在实际项目实践中有时候不希望某些服务调度到指定节点上，比如带GPU节点只运行带GPU需求的组件，其他不需要GPU服务不希望调度到GPU上。这种场景只需要在GPU节点设置一个污点。

场景1：k8s-node1不带GPU，k8s-node2带GPU，希望Pod可以调度到带GPU的节点

Step1：给k8s-node1节点打污点

[root@k8s-master 02]# kubectl taint node k8s-node1 gpu=no:NoSchedule
node/k8s-node1 tainted

1 2	[root@k8s-master 02]# kubectl taint node k8s-node1 gpu=no:NoSchedule node/k8s-node1 tainted

Step2：指定Pod创建时指定去 GPU 值为 "yes" 的节点

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

apiVersion: v1

kind: Pod

metadata:

labels:

run: pod-tolerations

spec:

containers:

- image: nginx

imagePullPolicy: IfNotPresent

tolerations:

- key: gpu

operator: Equal

value: "yes"

effect: NoSchedule

Step3：删除污点并更改 GPU 为 yes

[root@k8s-master 02]# kubectl taint node k8s-node1 gpu:NoSchedule-  
node/k8s-node1 untainted
[root@k8s-master 02]# kubectl taint node k8s-node1 gpu=yes:NoSchedule     
node/k8s-node1 tainted