Edit

Autoscale your AKS pods using Application Gateway Metrics (Beta)

  • Article

As incoming traffic increases, it becomes crucial to scale up your applications based on the demand.

In the following tutorial, we explain how you can use Application Gateway's AvgRequestCountPerHealthyHost metric to scale up your application. AvgRequestCountPerHealthyHost measures average requests sent to a specific backend pool and backend HTTP setting combination.

Use following two components:

  • Azure Kubernetes Metric Adapter - We use the metric adapter to expose Application Gateway metrics through the metric server. The Azure Kubernetes Metric Adapter is an open source project under Azure, similar to the Application Gateway Ingress Controller.
  • Horizontal Pod Autoscaler - We use HPA to use Application Gateway metrics and target a deployment for scaling.

Note

The Azure Kubernetes Metrics Adapter is no longer maintained. Kubernetes Event-driven Autoscaling (KEDA) is an alternative.
Also see Application Gateway for Containers.

Setting up Azure Kubernetes Metric Adapter

  1. First, create a Microsoft Entra service principal and assign it Monitoring Reader access over Application Gateway's resource group.

        applicationGatewayGroupName="<application-gateway-group-id>"
    applicationGatewayGroupId=$(az group show -g $applicationGatewayGroupName -o tsv --query "id")
    az ad sp create-for-rbac -n "azure-k8s-metric-adapter-sp" --role "Monitoring Reader" --scopes applicationGatewayGroupId

  2. Now, deploy the Azure Kubernetes Metric Adapter using the Microsoft Entra service principal created previously.

    kubectl create namespace custom-metrics
# use values from service principal created previously to create secret
kubectl create secret generic azure-k8s-metrics-adapter -n custom-metrics \
    --from-literal=azure-tenant-id=<tenantid> \
    --from-literal=azure-client-id=<clientid> \
    --from-literal=azure-client-secret=<secret>
kubectl apply -f kubectl apply -f https://raw.githubusercontent.com/Azure/azure-k8s-metrics-adapter/master/deploy/adapter.yaml -n custom-metrics

  3. Create an ExternalMetric resource with name appgw-request-count-metric. This resource instructs the metric adapter to expose AvgRequestCountPerHealthyHost metric for myApplicationGateway resource in myResourceGroup resource group. You can use the filter field to target a specific backend pool and backend HTTP setting in the Application Gateway.

    apiVersion: azure.com/v1alpha2
kind: ExternalMetric
metadata:
name: appgw-request-count-metric
spec:
    type: azuremonitor
    azure:
        resourceGroup: myResourceGroup # replace with your application gateway's resource group name
        resourceName: myApplicationGateway # replace with your application gateway's name
        resourceProviderNamespace: Microsoft.Network
        resourceType: applicationGateways
    metric:
        metricName: AvgRequestCountPerHealthyHost
        aggregation: Average
        filter: BackendSettingsPool eq '<backend-pool-name>~<backend-http-setting-name>' # optional

You can now make a request to the metric server to see if our new metric is getting exposed:

kubectl get --raw "/apis/external.metrics.k8s.io/v1beta1/namespaces/default/appgw-request-count-metric"
# Sample Output
# {
#   "kind": "ExternalMetricValueList",
#   "apiVersion": "external.metrics.k8s.io/v1beta1",
#   "metadata":
#     {
#       "selfLink": "/apis/external.metrics.k8s.io/v1beta1/namespaces/default/appgw-request-count-metric",
#     },
#   "items":
#     [
#       {
#         "metricName": "appgw-request-count-metric",
#         "metricLabels": null,
#         "timestamp": "2019-11-05T00:18:51Z",
#         "value": "30",
#       },
#     ],
# }

Using the new metric to scale up the deployment

Once we're able to expose appgw-request-count-metric through the metric server, we're ready to use Horizontal Pod Autoscaler to scale up our target deployment.

In following example, we target a sample deployment aspnet. We scale up Pods when appgw-request-count-metric > 200 per Pod up to a max of 10 Pods.

Replace your target deployment name and apply the following auto scale configuration:

apiVersion: autoscaling/v2beta1
kind: HorizontalPodAutoscaler
metadata:
  name: deployment-scaler
spec:
  scaleTargetRef:
    apiVersion: networking.k8s.io/v1
    kind: Deployment
    name: aspnet # replace with your deployment's name
  minReplicas: 1
  maxReplicas: 10
  metrics:
  - type: External
    external:
      metricName: appgw-request-count-metric
      targetAverageValue: 200

Test your setup by using a load test tool like apache bench:

ab -n10000 http://<applicaiton-gateway-ip-address>/

Next steps