Introduction

This is a sample of showing how to autoScale pods with the CPU usage.

Require

First of all, HPA depends on the metrics-server. If you did not deploy it already, please refer to the monitor sample repo.
Second, if we want to trigger the hpa feature with the CPU usage, we have to define the limit resource of CPU.

Here are three ways to define the cpu limitation:

limitRange-container.yaml

This is the definition for each container.

limitRange-pod.yaml

This is the definition for each pod. Each pod may contain more than one container.

deployment.yaml

Directly defined in the pod template as follows:

    spec:
      containers:
        - name: hpa-service
          image: local-dtr.com/kubernetes-hpa-sample:Develop
          ports:
            - containerPort: 8080
              name: http
          resources:
            limits:
              cpu: 200m
            requests:
              cpu: 101m

Deploy

mvn clean install && kubectl create -f ./kubernetes/

Test

Check metric-server

Get the cpu and memory usage of all pods:

$ kubectl top pods -n hpa-sample
NAME                      CPU(cores)   MEMORY(bytes)
hpa-dm-69cb57c8d4-gxgf8   200m         204Mi
hpa-dm-69cb57c8d4-mhfmz   200m         260Mi
hpa-dm-69cb57c8d4-rtlxq   200m         212Mi

When the java applications are starting, they use 100% cpu resources. We know that CPU is the compressible resource. We had setting the limitation of the CPU usage less than or equal to 200m before then. So that means our Settings are working.When When the java applications already started, watch the cpu usage:

$ kubectl top pods -n hpa-sample
NAME                      CPU(cores)   MEMORY(bytes)
hpa-dm-69cb57c8d4-gxgf8   2m           304Mi
hpa-dm-69cb57c8d4-mhfmz   2m           330Mi
hpa-dm-69cb57c8d4-rtlxq   2m           299Mi

They use a bit of the CPU resources and more memory resource.

Check endpoint

Here are three pods, they exposed port with the service resource which using the LoadBalancer type. Try many time:

$ curl 127.0.0.1:8880/hello
Hello, i am hpa-dm-69cb57c8d4-mhfmz.
$ curl 127.0.0.1:8880/hello
Hello, i am hpa-dm-69cb57c8d4-rtlxq.
$ curl 127.0.0.1:8880/hello
Hello, i am hpa-dm-69cb57c8d4-mhfmz.
$ curl 127.0.0.1:8880/hello
Hello, i am hpa-dm-69cb57c8d4-gxgf8.

Setting autoscale

Execute kubectl autoscale ** command as follows:

$ kubectl autoscale deployment hpa-dm --min=2 --max=5 --cpu-percent=15 -n hpa-sample

At the same time, open multiple screens to monitor the status of the deployment and events.

$ kubectl get events -n hpa-sample -w
LAST SEEN   TYPE     REASON                  OBJECT                         MESSAGE
18m         Normal   Created                 pod/hpa-dm-69cb57c8d4-rtlxq    Created container
18m         Normal   Started                 pod/hpa-dm-69cb57c8d4-rtlxq    Started container
18m         Normal   SuccessfulCreate        replicaset/hpa-dm-69cb57c8d4   Created pod: hpa-dm-69cb57c8d4-gxgf8
18m         Normal   SuccessfulCreate        replicaset/hpa-dm-69cb57c8d4   Created pod: hpa-dm-69cb57c8d4-mhfmz
18m         Normal   SuccessfulCreate        replicaset/hpa-dm-69cb57c8d4   Created pod: hpa-dm-69cb57c8d4-rtlxq
18m         Normal   ScalingReplicaSet       deployment/hpa-dm              Scaled up replica set hpa-dm-69cb57c8d4 to 3
0s          Normal   SuccessfulRescale       horizontalpodautoscaler/hpa-dm   New size: 2; reason: All metrics below target
0s          Normal   ScalingReplicaSet       deployment/hpa-dm                Scaled down replica set hpa-dm-69cb57c8d4 to 2
1s          Normal   SuccessfulDelete        replicaset/hpa-dm-69cb57c8d4     Deleted pod: hpa-dm-69cb57c8d4-gxgf8
0s          Normal   Killing                 pod/hpa-dm-69cb57c8d4-gxgf8      Killing container with id docker://hpa-service:Need to kill Pod

The last four lines show that the horizontalpodautoscaler triggered the auto scale capacity.

$ kubectl get deployment hpa-dm -n hpa-sample -w
NAME     READY   UP-TO-DATE   AVAILABLE   AGE
hpa-dm   3/3     3            3           24m
hpa-dm   3/2     3            3           28m
hpa-dm   3/2     3            3           28m
hpa-dm   2/2     2            2           28m

Final there are 2 available pods after finishing the auto scale. View the change of the hpa, the output shows the metrics value which it is monitoring.

$ kubectl get hpa  -w -n  hpa-sample
NAME     REFERENCE           TARGETS         MINPODS   MAXPODS   REPLICAS   AGE
hpa-dm   Deployment/hpa-dm   <unknown>/15%   2         5         0          0s
hpa-dm   Deployment/hpa-dm   1%/15%          2         5         3          30s
hpa-dm   Deployment/hpa-dm   1%/15%          2         5         2          60s

Through the kubectl get hpa -o yaml -n hpa-sample command, it will show how the hpa defined.

apiVersion: autoscaling/v1
kind: HorizontalPodAutoscaler
metadata:
  name: hpa-dm
  namespace: hpa-sample
spec:
  maxReplicas: 5
  minReplicas: 2
  scaleTargetRef:
    apiVersion: extensions/v1beta1
    kind: Deployment
    name: hpa-dm
  targetCPUUtilizationPercentage: 15

Above just show the main definition of hpa and delete useless information.

Others

In addition to using the HPA functionality on the command line, it can also be defined using yaml, see more.

$ kubectl apply -f ./kubernetes/hpa/hpa.yaml

Cleanup

$ kubectl delete namespace hpa-sample

SoulSong/kubernetes-hpa-sample