Horizontal Pod Autoscaler (HPA)

CPU

  • Deploy frontend app (if you still not deploy it yet)

    oc new-project project1
oc apply -f manifests/backend-v1.yaml -n project1
oc apply -f manifests/backend-service.yaml -n project1
oc apply -f manifests/backend-route.yaml -n project1
oc set env deployment backend-v1 -n project1 APP_BACKEND=https://httpbin.org/status/200
oc wait --for=condition=ready --timeout=60s pod -l app=backend
BACKEND_URL=http://$(oc get route backend -n project1 -o jsonpath='{.spec.host}')
curl -v -k $BACKEND_URL

  • Review CPU HPA for deployment backend

    • Scale out when average CPU utilization is greater than 20% of CPU limit
    • Maximum pods is 5

    • Scale down to min replicas if utilization is lower than threshold for 60 sec

        minReplicas: 1
  maxReplicas: 5
  targetCPUUtilizationPercentage: 20
  behavior:
    scaleDown: 
      policies: 
      - type: Pods 
        value: 1
        periodSeconds: 60 
      selectPolicy: Min 
      stabilizationWindowSeconds: 70 
    scaleUp: 
      policies:
      - type: Pods
        value: 5
        periodSeconds: 70
      selectPolicy: Max
      stabilizationWindowSeconds: 0

  • Create CPU HPA for deployment backend

oc apply -f manifests/backend-cpu-hpa.yaml -n project1
  • Check HPA status
    watch oc get horizontalpodautoscaler/backend-cpu -n project1

Output

NAME          REFERENCE            TARGETS         MINPODS   MAXPODS   REPLICAS   AGE
backend-cpu   Deployment/backend   <unknown>/20%   1         5         0          2m4s

  • Load test with K6

    • 50 threads
    • Duration 3 minutes
    • Ramp up 30 sec

    • Ramp down 30 sec

      URL=http://backend.project1:8080
oc run load-test -n project1 -i \
--image=loadimpact/k6 --rm=true --restart=Never \
--  run -  < manifests/load-test-k6.js \
-e URL=$URL -e THREADS=50 -e DURATION=3m -e RAMPUP=30s -e RAMPDOWN=30s

      Output

      Check pods CPU in developer console

  • Stop load test and wait 1 minutes to scale down to 1 replica.

    Remark: k6 will run as pod name load-test for 4 minutes if you want to force stop before 4 minutes just delete load-test pod

      oc delete pod load-test -n project1

    Memory

  • Deploy Memory Leak App

          resources:
        requests:
          cpu: "0.1"
          memory: 100Mi
        limits:
          cpu: "0.2"
          memory: 200Mi
      env:
        - name: APP_DISHSIZE
          value: "10485760"

    Deploy Leak App

    oc apply -f manifests/leak.yaml -n project1
oc expose deployment leak -n project1
oc expose svc leak -n project1

    Check for leak app memory limit

  • Review Memory HPA for deployment leak app

    • Scale out when average memory utilization is greater than 100Mi
    • Maximum pods is 3

    • Scale down to min replicas if utilization is lower than threshold for 60 sec

  • Create Memory HPA for deployment frontend v1

    oc apply -f manifests/leak-memory-hpa.yaml -n project1

  • Check HPA status

    watch oc get horizontalpodautoscaler/leak-memory -n project1

    Output

    NAME          REFERENCE         TARGETS          MINPODS   MAXPODS   REPLICAS   AGE
leak-memory   Deployment/leak   81829888/100Mi   1         3         1          5h37m

  • Test Leak App to consume another 30 MB

    curl -v http://$(oc get route leak -n project1 -o jsonpath='{.spec.host}')/eat/3
watch oc get po -l app=leak -n project1

    Output

    < HTTP/1.1 200 OK
< content-type: text/plain;charset=UTF-8
< content-length: 20
< set-cookie: 5ba5938c43542864828d859a0c1ca92a=00adce54797d4ef2002e330336942b56; path=/; HttpOnly
< cache-control: private
<
* Connection #0 to host leak-project1.apps.cluster-27h8w.27h8w.sandbox74.opentlc.com left intact
Added 31457280 bytes

  • Check that another leak pod is created

    NAME                    READY   STATUS    RESTARTS   AGE
leak-74d885844d-s4vjz   1/1     Running   0          26s
leak-74d885844d-vrx6s   1/1     Running   0          2m45s

  • Check HPA status again

    NAME          REFERENCE         TARGETS           MINPODS   MAXPODS   REPLICAS   AGE
leak-memory   Deployment/leak   127311872/100Mi   1         3         2          97s

  • Check memory utilization from Developer console

Custom Metrics Autoscaler

Configure User Workload Monitoring

  • Enable user workload monitoring

  • Create Service Monitoring to monitor backend service

    oc apply -f manifests/backend-service-monitor.yaml -n project1

  • Remove HPA by CPU and run K6 with 10 threads

      oc delete hpa/backend-cpu -n project1
  URL=http://backend.project1:8080
  oc run load-test -n project1 -i \
  --image=loadimpact/k6 --rm=true --restart=Never \
  --  run -  < manifests/load-test-k6.js \
  -e URL=$URL -e THREADS=10 -e DURATION=3m -e RAMPUP=30s -e RAMPDOWN=30s

  • Check for backends's concurrent requests/sec metric

    • Open Developer Console -> Observe -> Metrics -> Custom metrics

    • Input following PromQL to check for number of requests/min

      rate(http_server_requests_seconds_count{method="GET",uri="root",outcome="SUCCESS"}[1m])

      Output

Install Operator

  • Install Custom Metrics Autoscaler Operator

    Verify 

    oc get csv -n openshift-keda
oc get po -n openshift-keda

    Output

    NAME                                    DISPLAY                     VERSION      REPLACES   PHASE
custom-metrics-autoscaler.v2.10.1-253   Custom Metrics Autoscaler   2.10.1-253              Succeeded
NAME                                                 READY   STATUS    RESTARTS   AGE
custom-metrics-autoscaler-operator-9c9c7c4cb-26l28   1/1     Running   0          10m

Create SacledObject

  • Create Service Account 

    oc create serviceaccount thanos -n project1

  • Create TriggerAuthentication and assign cluster role to query thanos

    TOKEN=$(oc describe serviceaccount thanos -n project1|grep Token|awk -F':' '{print $2}'|awk '$1=$1')
cat manifests/cma-trigger-authentication.yaml|sed 's/TOKEN/'$TOKEN'/'|sed 's/PROJECT/project1/' | oc create -f -
oc adm policy add-cluster-role-to-user cluster-monitoring-view -z thanos -n project1

  • Create ScaledObject to scale backend pod by conncurrent request.

    oc apply -f manifests/backend-scaled-object.yaml -n project1
oc get scaledobject/backend -o yaml -n project1 | grep -A4 " conditions:"

    Number of replicas and triggers configuration

    triggers:
    - authenticationRef:
        kind: TriggerAuthentication
        name: keda-trigger-auth-prometheus
      metadata:
        metricName: http_server_requests_seconds_count
        namespace: project1
        query: sum(rate(http_server_requests_seconds_count{method="GET", uri="root", outcome="SUCCESS"}[1m])) by (pod)
        serverAddress: https://thanos-querier.openshift-monitoring.svc.cluster.local:9092
        threshold: "15"
        authModes: "bearer"

    Output

     scaledobject.keda.sh/backend created
 conditions:
 - message: ScaledObject is defined correctly and is ready for scaling
   reason: ScaledObjectReady
   status: "True"
   type: Ready

Test

  • Run Load Test with K6

      URL=http://$(oc get route backend -n project1 -o jsonpath='{.spec.host}')
  oc run load-test -n project1 -i \
  --image=loadimpact/k6 --rm=true --restart=Never \
  --  run -  < manifests/load-test-k6.js \
  -e URL=$URL -e THREADS=20 -e DURATION=3m -e RAMPUP=30s -e RAMPDOWN=30s

  • Check that ScaledObject is active

    NAME      SCALETARGETKIND      SCALETARGETNAME   MIN   MAX   TRIGGERS     AUTHENTICATION                 READY   ACTIVE   FALLBACK   AGE
backend   apps/v1.Deployment   backend-v1        1     5     prometheus   keda-trigger-auth-prometheus   True    True     False      77m

  • Check that backend is scaled out

    NAME                         READY   STATUS    RESTARTS   AGE
backend-v1-c448b55d9-5t2j6   1/1     Running   0          13s
backend-v1-c448b55d9-5v9gm   1/1     Running   0          129m

  • Check concurent request. Open Developer Console and use custom query

    sum by (pod)(rate(http_server_requests_seconds_count{method="GET",uri="root",outcome="SUCCESS"}[1m]))

results matching ""

    No results matching ""