Kubernetes使用Keda进行d性伸缩，更合理利用资源

我最新最全的文章都在 南瓜慢说 www.pkslow.com ，欢迎大家来喝茶！

1 简介

Kubernetes自带的HPA是只支持CPU/MEM的，很多时候我们并不根据这两项指标来进行伸缩资源。比如消费者不断处理MQ的消息，我们希望MQ如果堆积过多，就启动更多的消费者来处理任务。而Keda给了我们很多选择。

KEDA 是 Kubernetes 基于事件驱动的自动伸缩工具，通过 KEDA 我们可以根据需要处理的事件数量来驱动 Kubernetes 中任何容器的扩展。KEDA 可以直接部署到任何 Kubernetes 集群中和标准的组件一起工作。

Keda所支持的事件源非常丰富，本文我们以RabbitMQ为例进行演示。

2 安装Keda

安装的方法很多，我们直接通过yaml文件来安装，这样还可以修改镜像地址等。先从（ https://github.com/kedacore/keda/releases/download/v2.2.0/keda-2.2.0.yaml ）下载yaml文件，然后执行：

$ kubectl apply -f ~/Downloads/keda-2.2.0.yaml namespace/keda created customresourcedefinition.apiextensions.k8s.io/clustertriggerauthentications.keda.sh created customresourcedefinition.apiextensions.k8s.io/scaledjobs.keda.sh created customresourcedefinition.apiextensions.k8s.io/scaledobjects.keda.sh created customresourcedefinition.apiextensions.k8s.io/triggerauthentications.keda.sh created serviceaccount/keda-operator created clusterrole.rbac.authorization.k8s.io/keda-external-metrics-reader created clusterrole.rbac.authorization.k8s.io/keda-operator created rolebinding.rbac.authorization.k8s.io/keda-auth-reader created clusterrolebinding.rbac.authorization.k8s.io/keda-hpa-controller-external-metrics created clusterrolebinding.rbac.authorization.k8s.io/keda-operator created clusterrolebinding.rbac.authorization.k8s.io/keda:system:auth-delegator created service/keda-metrics-apiserver created deployment.apps/keda-metrics-apiserver created deployment.apps/keda-operator created apiservice.apiregistration.k8s.io/v1beta1.external.metrics.k8s.io created

检查一下是否都已经启动完成：

$ kubectl get all -n keda NAME READY STATUS RESTARTS AGE pod/keda-metrics-apiserver-55dc9f9498-smc2d 1/1 Running 0 2m41s pod/keda-operator-59dcf989d6-pxcbb 1/1 Running 0 2m41s NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE service/keda-metrics-apiserver ClusterIP 10.104.255.44 <none> 443/TCP,80/TCP 2m41s NAME READY UP-TO-DATE AVAILABLE AGE deployment.apps/keda-metrics-apiserver 1/1 1 1 2m42s deployment.apps/keda-operator 1/1 1 1 2m42s NAME DESIRED CURRENT READY AGE replicaset.apps/keda-metrics-apiserver-55dc9f9498 1 1 1 2m42s replicaset.apps/keda-operator-59dcf989d6 1 1 1 2m42s

也可以看到镜像多了：

$ docker images | grep keda ghcr.io/kedacore/keda-metrics-apiserver 2.2.0 a43d40453368 6 weeks ago 95.3MB ghcr.io/kedacore/keda 2.2.0 42b88f042914 6 weeks ago 83MB

如果要卸载请执行：

$ kubectl delete -f ~/Downloads/keda-2.2.0.yaml 3 安装RabbitMQ

为了快速安装，也方便日后删除，我们通过Helm来安装RabbitMQ。

查看可用的chart：

$ helm search repo rabbit

执行安装：

$ helm install azure-rabbitmq azure/rabbitmq

检查一下：

$ helm list NAME NAMESPACE REVISION UPDATED STATUS CHART APP VERSION azure-ingress default 1 2021-02-14 01:21:07.212107 +0800 CST deployed nginx-ingress-1.41.3 v0.34.1 azure-rabbitmq default 1 2021-05-05 11:29:06.979437 +0800 CST deployed rabbitmq-6.18.2 3.8.2

用户名为user，密码获取如下：

$ echo "Password : $(kubectl get secret --namespace default azure-rabbitmq -o jsonpath="{.data.rabbitmq-password}" | base64 --decode)" Password : YNsEayx8w2 4 测试

部署消费者，注意这里有个MQ连接信息和加密，要根据自己情况修改。

$ kubectl apply -f src/main/kubernetes/deploy-consumer.yaml secret/rabbitmq-consumer-secret created deployment.apps/rabbitmq-consumer created scaledobject.keda.sh/rabbitmq-consumer created triggerauthentication.keda.sh/rabbitmq-consumer-trigger created

查看deployment，发现是没有Pod创建，因为还不需要处理，MQ现在的队列为0。

$ kubectl get deployments NAME READY UP-TO-DATE AVAILABLE AGE azure-ingress-nginx-ingress-controller 1/1 1 1 80d azure-ingress-nginx-ingress-default-backend 1/1 1 1 80d rabbitmq-consumer 0/0 0 0 131m

部署生产者，往MQ发送消息：

$ kubectl apply -f src/main/kubernetes/deploy-publisher-job.yaml job.batch/rabbitmq-publish created

可以看到，慢慢消费者就起来了，并且创建了越来越多的Pod来处理MQ：

$ kubectl get deployments rabbitmq-consumer NAME READY UP-TO-DATE AVAILABLE AGE rabbitmq-consumer 1/1 1 1 167m $ kubectl get deployments rabbitmq-consumer NAME READY UP-TO-DATE AVAILABLE AGE rabbitmq-consumer 3/4 4 3 168m $ kubectl get deployments rabbitmq-consumer NAME READY UP-TO-DATE AVAILABLE AGE rabbitmq-consumer 4/8 8 4 168m $ kubectl get deployments rabbitmq-consumer NAME READY UP-TO-DATE AVAILABLE AGE rabbitmq-consumer 6/8 8 6 169m $ kubectl get deployments rabbitmq-consumer NAME READY UP-TO-DATE AVAILABLE AGE rabbitmq-consumer 0/0 0 0 171m

查看Deployment的Event也可以看到结果：

Events: Type Reason Age From Message ---- ------ ---- ---- ------- Normal ScalingReplicaSet 5m55s (x2 over 172m) deployment-controller Scaled up replica set rabbitmq-consumer-7b477f78b4 to 1 Normal ScalingReplicaSet 5m6s deployment-controller Scaled up replica set rabbitmq-consumer-7b477f78b4 to 4 Normal ScalingReplicaSet 4m6s deployment-controller Scaled up replica set rabbitmq-consumer-7b477f78b4 to 8 Normal ScalingReplicaSet 3m5s deployment-controller Scaled up replica set rabbitmq-consumer-7b477f78b4 to 16 Normal ScalingReplicaSet 3m3s (x2 over 172m) deployment-controller Scaled down replica set rabbitmq-consumer-7b477f78b4 to 0

处理完成后，又会回到0了。

总结

代码请查看：https://github.com/LarryDpk/pkslow-samples

---

欢迎关注微信公众号<南瓜慢说>，将持续为你更新...

多读书，多分享；多写作，多整理。

Kubernetes使用Keda进行d性伸缩，更合理利用资源