有关EKS系列文档和实验的汇总，请点击这里跳转
前文介绍了为NodeGroup的扩容和更换机型，但使用的都是On-Demand机型（对应RI预留实例计费方式），本文介绍为NodeGroup使用Spot实例。
一、使用Spot实例
1、使用ec2-instance-selector查看可选合适的机型
要查看能提供特定CPU和内存比例数量的机型，可以使用ec2-instance-selector工具，参考这个文章。
执行如下命令可以查看相似进行：
ec2-instance-selector --base-instance-type m5.2xlarge --flexible也可以手工指定vCPU和内存进行查询
ec2-instance-selector --vcpus 8 --memory 32即可获得匹配的机型清单。
2、新建使用新的EC2规格的nodegroup
编辑如下内容，并保存为spot-nodegroup.yaml文件。需要注意的是，如果新创建的Nodegroup在Public Subnet内，这直接使用如下内容即可。如果新创建的Nodegroup在Private Subnet内，那么请增加如下一行privateNetworking: true到配置文件中。添加的位置在volumeSize的下一行即可。
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig
metadata:
name: eksworkshop
region: ap-southeast-1
version: "1.23"
managedNodeGroups:
- name: spot1
labels:
Name: spot1
instanceTypes: ["m4.2xlarge","m5.2xlarge","m5d.2xlarge"]
spot: true
minSize: 3
desiredCapacity: 3
maxSize: 6
volumeType: gp3
volumeSize: 100
tags:
nodegroup-name: spot1
iam:
withAddonPolicies:
imageBuilder: true
autoScaler: true
certManager: true
efs: true
albIngress: true
xRay: true
cloudWatch: true编辑完毕后保存退出。执行如下命令创建：
eksctl create nodegroup -f spot-nodegroup.yaml执行如下命令查看nodegroup是否正常。请注意修改cluster名称与当前集群名称要匹配。
eksctl get nodegroup --cluster eksworkshop返回结果可以看到原来的和新创建的两个nodegroup如下。
CLUSTER NODEGROUP STATUS CREATED MIN SIZE MAX SIZE DESIRED CAPACITY INSTANCE TYPE IMAGE ID ASG NAME TYPE
eksworkshop managed-ng ACTIVE 2022-10-14T08:25:52Z 3 6 3 m5.2xlarge AL2_x86_64 eks-managed-ng-32c1eacc-bd74-6172-37c3-d9a66afa8512 managed
eksworkshop spot1 ACTIVE 2022-10-14T09:39:29Z 3 6 3 m4.2xlarge,m5.2xlarge,m5d.2xlarge AL2_x86_64 eks-spot1-88c1eaee-6f0f-0da3-a6a4-3b2792ee447a managed查看节点的创建时间：
kubectl get nodes --sort-by=.metadata.creationTimestamp返回信息类似如下：
NAME STATUS ROLES AGE VERSION
ip-192-168-73-49.ap-southeast-1.compute.internal Ready <none> 3h30m v1.23.9-eks-ba74326
ip-192-168-3-225.ap-southeast-1.compute.internal Ready <none> 3h30m v1.23.9-eks-ba74326
ip-192-168-53-33.ap-southeast-1.compute.internal Ready <none> 3h30m v1.23.9-eks-ba74326
ip-192-168-27-19.ap-southeast-1.compute.internal Ready <none> 136m v1.23.9-eks-ba74326
ip-192-168-46-69.ap-southeast-1.compute.internal Ready <none> 136m v1.23.9-eks-ba74326
ip-192-168-65-14.ap-southeast-1.compute.internal Ready <none> 136m v1.23.9-eks-ba743263、查看当前集群的On-Demand节点
执行如下命令：
kubectl get nodes \
--label-columns=eks.amazonaws.com/capacityType \
--selector=eks.amazonaws.com/capacityType=ON_DEMAND输出结果类似如下：
NAME STATUS ROLES AGE VERSION CAPACITYTYPE
ip-192-168-3-225.ap-southeast-1.compute.internal Ready <none> 68m v1.23.9-eks-ba74326 ON_DEMAND
ip-192-168-53-33.ap-southeast-1.compute.internal Ready <none> 68m v1.23.9-eks-ba74326 ON_DEMAND
ip-192-168-73-49.ap-southeast-1.compute.internal Ready <none> 68m v1.23.9-eks-ba74326 ON_DEMAND4、查看当前集群的Spot节点
查看节点类型是Spot的Node：
kubectl get nodes \
--label-columns=eks.amazonaws.com/capacityType \
--selector=eks.amazonaws.com/capacityType=SPOT返回结果如下：
NAME STATUS ROLES AGE VERSION CAPACITYTYPE
ip-192-168-27-19.ap-southeast-1.compute.internal Ready <none> 141m v1.23.9-eks-ba74326 SPOT
ip-192-168-46-69.ap-southeast-1.compute.internal Ready <none> 141m v1.23.9-eks-ba74326 SPOT
ip-192-168-65-14.ap-southeast-1.compute.internal Ready <none> 141m v1.23.9-eks-ba74326 SPOT二、部署应用分别运行在Spot和On-demand机型上
1、使用Spot的yaml配置文件
在和Container配置平级的位置，增加如下一段配置，表示优先使用Spot节点运行服务：
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
- key: eks.amazonaws.com/capacityType
operator: In
values:
- SPOT
tolerations:
- key: "spotInstance"
operator: "Equal"
value: "true"
effect: "PreferNoSchedule"在和Container配置平级的位置，增加如下一段配置，表示强制使用On-demand节点运行服务：
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: eks.amazonaws.com/capacityType
operator: In
values:
- ON_DEMAND2、使用Spot的yaml文件示例
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: demo-nginx-od
labels:
app: demo-nginx-od
spec:
replicas: 3
selector:
matchLabels:
app: demo-nginx-od
template:
metadata:
labels:
app: demo-nginx-od
spec:
affinity:
nodeAffinity:
preferredDuringSchedulingIgnoredDuringExecution:
- weight: 1
preference:
matchExpressions:
- key: eks.amazonaws.com/capacityType
operator: In
values:
- SPOT
tolerations:
- key: "spotInstance"
operator: "Equal"
value: "true"
effect: "PreferNoSchedule"
containers:
- name: demo-nginx-od
image: public.ecr.aws/nginx/nginx:1.21.6-alpine
ports:
- containerPort: 80
resources:
limits:
cpu: "1"
memory: 2G
---
apiVersion: v1
kind: Service
metadata:
name: "demo-nginx-od"
annotations:
service.beta.kubernetes.io/aws-load-balancer-type: nlb
spec:
selector:
app: demo-nginx-od
type: LoadBalancer
ports:
- protocol: TCP
port: 80
targetPort: 803、使用On-demand的yaml文件示例
---
apiVersion: apps/v1
kind: Deployment
metadata:
name: demo-nginx-spot
labels:
app: demo-nginx-spot
spec:
replicas: 3
selector:
matchLabels:
app: demo-nginx-spot
template:
metadata:
labels:
app: demo-nginx-spot
spec:
affinity:
nodeAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
nodeSelectorTerms:
- matchExpressions:
- key: eks.amazonaws.com/capacityType
operator: In
values:
- ON_DEMAND
containers:
- name: demo-nginx-spot
image: public.ecr.aws/nginx/nginx:1.21.6-alpine
ports:
- containerPort: 80
resources:
limits:
cpu: "1"
memory: 2G
---
apiVersion: v1
kind: Service
metadata:
name: "demo-nginx-spot"
annotations:
service.beta.kubernetes.io/aws-load-balancer-type: nlb
spec:
selector:
app: demo-nginx-spot
type: LoadBalancer
ports:
- protocol: TCP
port: 80
targetPort: 80三、查看在On-demand和Spot节点上的Pod
1、查看Spot节点上的Pod
执行如下命令：
for n in $(kubectl get nodes -l eks.amazonaws.com/capacityType=SPOT --no-headers | cut -d " " -f1); do echo "Pods on instance ${n}:";kubectl get pods --all-namespaces --no-headers --field-selector spec.nodeName=${n} ; echo ; done返回信息如下：
Pods on instance ip-192-168-27-19.ap-southeast-1.compute.internal:
default demo-nginx-spot-79676dcc79-2mlv8 1/1 Running 0 117s
kube-system aws-node-pdcx2 1/1 Running 1 (11h ago) 2d3h
kube-system kube-proxy-j5xdm 1/1 Running 1 (11h ago) 2d3h
Pods on instance ip-192-168-46-69.ap-southeast-1.compute.internal:
default demo-nginx-spot-79676dcc79-gpk4z 1/1 Running 0 118s
kube-system aws-node-nmz6k 1/1 Running 1 (11h ago) 2d3h
kube-system kube-proxy-qpb6f 1/1 Running 1 (11h ago) 2d3h
Pods on instance ip-192-168-65-14.ap-southeast-1.compute.internal:
default demo-nginx-spot-79676dcc79-bllnb 1/1 Running 0 119s
kube-system aws-node-vbl6f 1/1 Running 1 (11h ago) 2d3h
kube-system kube-proxy-45jfj 1/1 Running 1 (11h ago) 2d3h2、查看On-demand节点上的Pod
执行如下命令：
for n in $(kubectl get nodes -l eks.amazonaws.com/capacityType=ON_DEMAND --no-headers | cut -d " " -f1); do echo "Pods on instance ${n}:";kubectl get pods --all-namespaces --no-headers --field-selector spec.nodeName=${n} ; echo ; done返回信息如下：
Pods on instance ip-192-168-3-225.ap-southeast-1.compute.internal:
default demo-nginx-on-demand-5549798b4c-ftwq2 1/1 Running 0 3m5s
kube-system aws-node-lgwn5 1/1 Running 1 (11h ago) 2d4h
kube-system kube-proxy-kvqd9 1/1 Running 1 (11h ago) 2d4h
Pods on instance ip-192-168-53-33.ap-southeast-1.compute.internal:
default demo-nginx-on-demand-5549798b4c-zkns8 1/1 Running 0 3m6s
kube-system aws-node-jw6t2 1/1 Running 1 (11h ago) 2d4h
kube-system kube-proxy-6ncrr 1/1 Running 1 (11h ago) 2d4h
Pods on instance ip-192-168-73-49.ap-southeast-1.compute.internal:
default demo-nginx-on-demand-5549798b4c-nbxf8 1/1 Running 0 3m7s
kube-system aws-node-7wh29 1/1 Running 1 (11h ago) 2d4h
kube-system coredns-6d8cc4bb5d-bfmmb 1/1 Running 1 (11h ago) 2d5h
kube-system coredns-6d8cc4bb5d-fhj8p 1/1 Running 1 (11h ago) 2d5h
kube-system kube-proxy-2wm22 1/1 Running 1 (11h ago) 2d4h