Install ack-kubernetes-elastic-workload

You must resolve the following issues when you use Kubernetes to schedule pods and manage the lifecycle of pods:

• How to modify the pod scheduling policy when the number of replicated pods reaches a threshold.
• How to prioritize specific pods when Kubernetes manages the lifecycle of pods.

Create an application by using a Deployment based on the following template:

apiVersion: apps/v1 # for versions before 1.8.0 use apps/v1beta1
kind: Deployment
metadata:
  name: nginx-deployment-basic
  labels:
    app: nginx
spec:
  replicas: 2
  selector:
    matchLabels:
      app: nginx
  template:
    metadata:
      labels:
        app: nginx
    spec:
    #  nodeSelector:
    #    env: test-team
      containers:
      - name: nginx
        image: nginx:1.7.9 
        ports:
        - containerPort: 80

Create an elastic workload for the Deployment.

# Define an elastic workload. 
apiVersion: autoscaling.alibabacloud.com/v1beta1
kind: ElasticWorkload
metadata:
  name: elasticworkload-sample
spec:
  sourceTarget:
    name: nginx-deployment-basic
    kind: Deployment
    apiVersion: apps/v1
    min: 2
    max: 4
  replicas: 6
  elasticUnit:
  - name: virtual-kubelet
    labels:
      alibabacloud.com/eci: "true"
    # min: 0 You can specify the maximum and minimum numbers of replicated pods for each elastic unit. 
    # max: 10

Typically, the configurations of an elastic workload include the following parts:

1. SourceTarget: defines the type of the source workload, and the maximum and minimum numbers of replicated pods.
2. elasticUnit: an array that defines the scheduling policy for an elastic unit. To define scheduling policies for multiple elastic units, specify related parameters in the order shown in the template.

In this example, the number of replicated pods provisioned for the source workload ranges from 2 to 4. If the elastic workload has 2 to 4 replicated pods, all the pods are scheduled to the source workload. If the elastic workload contains more than 4 pods, the remaining pods are scheduled to the virtual-kubelet elastic unit. Replicated pods are scheduled based on the scheduling policy of virtual-kubelet, such as the label, annotation, nodeSelector, affinity, and toleration parameters.

Run the following command to query the status of the elastic workload:

kubectl describe ew elasticworkload-sample   # same as kubectl get elasticworkload

Expected output:

Name:         elasticworkload-sample
Namespace:    default
Labels:       <none>
Annotations:  <none>
API Version:  autoscaling.alibabacloud.com/v1beta1
Kind:         ElasticWorkload
Metadata:
  Creation Timestamp:  2021-04-30T06:58:03Z
  Generation:          2
  Resource Version:  1126368056
  Self Link:         /apis/autoscaling.alibabacloud.com/v1beta1/namespaces/default/elasticworkloads/elasticworkload-sample
  UID:               a465de9e-1253-4cbe-8cd5-98393393e990
Spec:
  Elastic Unit:
    Labels:
      alibabacloud.com/eci:  true
    Name:                    virtual-kubelet
  Replicas:                  6
  Source Target:
    API Version:  apps/v1
    Kind:         Deployment
    Max:          4
    Min:          2
    Name:         nginx-deployment-basic
Status:
  Elastic Units Status:
    Desired Replicas:  2
    Name:              nginx-deployment-basic-unit-virtual-kubelet
    Update Timestamp:  2021-04-30T06:58:03Z
  Replicas:            6
  Selector:            app=nginx
  Source Target:
    API Version:       apps/v1
    Desired Replicas:  4
    Kind:              Deployment
    Name:              nginx-deployment-basic
    Update Timestamp:  2021-04-30T06:58:03Z
Events:
  Type    Reason                 Age                  From             Message
  ----    ------                 ----                 ----             -------
  Normal  UnitCreation           110s                 ElasticWorkload  ElasticWorkloadUnit nginx-deployment-basic-unit-virtual-kubelet created
  Normal  ElasticWorkloadUpdate  110s (x2 over 110s)  ElasticWorkload  ElasticWorkload update
  Normal  UnitUpdate             110s                 ElasticWorkload  ElasticWorkloadUnit virtual-kubelet has been updated

Run the following command to query the status of the pods:

kubectl get pod -o wide

Expected output:

NAME                                                           READY   STATUS    RESTARTS   AGE   IP              NODE                            NOMINATED NODE   READINESS GATES
nginx-deployment-basic-769f84b5cf-hdmlw                        1/1     Running   0          46m   172.26.240.69   cn-hangzhou.10.1.84.111         <none>           <none>
nginx-deployment-basic-769f84b5cf-lmd99                        1/1     Running   0          46m   172.26.240.6    cn-hangzhou.10.1.84.112         <none>           <none>
nginx-deployment-basic-769f84b5cf-nbp5c                        1/1     Running   0          30m   172.26.240.75   cn-hangzhou.10.1.84.111         <none>           <none>
nginx-deployment-basic-769f84b5cf-scj68                        1/1     Running   0          30m   172.26.240.11   cn-hangzhou.10.1.84.112         <none>           <none>
nginx-deployment-basic-unit-virtual-kubelet-594f86b5c9-8z876   1/1     Running   0          10m   10.1.84.119     virtual-kubelet-cn-hangzhou-i   <none>           <none>
nginx-deployment-basic-unit-virtual-kubelet-594f86b5c9-drxvq   1/1     Running   0          10m   10.1.84.118     virtual-kubelet-cn-hangzhou-i   <none>           <none>

You can also use HPA together with ack-kubernetes-elastic-workload to scale an elastic workload, as shown in the preceding figure. If a Kubernetes workload is scaled by HPA, ack-kubernetes-elastic-workload adjusts the number of replicated pods scheduled to each elastic unit. For example, if HPA reduces the number of replicated pods in an elastic workload from 6 to 4, ack-kubernetes-elastic-workload preferably reduces the replicated pods of elastic units instead of reducing the number of replicated pods in the source workload.

apiVersion: autoscaling/v2beta2
kind: HorizontalPodAutoscaler
metadata:
  name: elastic-workload-demo
  namespace: default
spec:
  scaleTargetRef:
    apiVersion: autoscaling.alibabacloud.com/v1beta1
    kind: ElasticWorkload
    name: elasticworkload-sample
  minReplicas: 2
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 50