KEDA HTTP Add-on
Getting Started
Concepts
User Guide
Operations
Reference

Getting Started Latest

Sample HTTP application deployment and autoscaling with the KEDA HTTP Add-on

Suggest a change

In this tutorial, we will install the KEDA HTTP Add-on and use it to autoscale an HTTP application based on incoming traffic — including scaling to zero when idle.

By the end, we will have:

  • A sample HTTP application running in our cluster
  • The HTTP Add-on intercepting and counting requests
  • KEDA scaling the application up under load and back to zero when traffic stops

Prerequisites

Before we begin, we need:

  • A Kubernetes cluster (kind, minikube, or a cloud provider)

  • kubectl configured to access the cluster

  • Helm 3 installed

  • KEDA core installed:

    helm install keda kedacore/keda --namespace keda --create-namespace

    See the KEDA deployment docs for other installation methods.

Step 1: Add the KEDA Helm Repository

If we have not already added the KEDA Helm repository, we add it now and update our local chart index:

helm repo add kedacore https://kedacore.github.io/charts
helm repo update

Step 2: Install the HTTP Add-on

We install the HTTP Add-on into the same keda namespace where KEDA core is running:

helm install http-add-on kedacore/keda-add-ons-http --namespace keda

We verify that all components are running:

kubectl get pods -n keda

We will see pods for the operator, interceptor, and scaler — all with a Running status:

NAME                                                   READY   STATUS    RESTARTS   AGE
keda-add-ons-http-interceptor-...                      1/1     Running   0          30s
keda-add-ons-http-operator-...                         1/1     Running   0          30s
keda-add-ons-http-scaler-...                           1/1     Running   0          30s
keda-admission-webhooks-...                            1/1     Running   0          2m
keda-operator-...                                      1/1     Running   0          2m
keda-operator-metrics-apiserver-...                    1/1     Running   0          2m

Step 3: Deploy a Sample Application

We create a namespace and deploy a sample HTTP application using traefik/whoami, a lightweight HTTP server that responds with request metadata.

kubectl create namespace demo

We deploy a Deployment and Service:

kubectl apply -f - <<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: sample-app
  namespace: demo
spec:
  selector:
    matchLabels:
      app: sample-app
  template:
    metadata:
      labels:
        app: sample-app
    spec:
      containers:
        - name: sample-app
          image: traefik/whoami
          args: ["--port=8080"]
          ports:
            - containerPort: 8080
---
apiVersion: v1
kind: Service
metadata:
  name: sample-app
  namespace: demo
spec:
  selector:
    app: sample-app
  ports:
    - port: 80
      targetPort: 8080
EOF

We verify the Deployment was created:

kubectl get deployment -n demo

We will see the Deployment with 1 replica running (the Kubernetes default):

NAME         READY   UP-TO-DATE   AVAILABLE   AGE
sample-app   1/1     1            1           10s

Step 4: Create an InterceptorRoute

The InterceptorRoute tells the interceptor how to route requests to our sample app and what scaling metric to use.

kubectl apply -f - <<EOF
apiVersion: http.keda.sh/v1beta1
kind: InterceptorRoute
metadata:
  name: sample-app
  namespace: demo
spec:
  target:
    service: sample-app
    port: 80
  rules:
    - hosts:
        - sample-app.example.com
  scalingMetric:
    requestRate:
      targetValue: 5
      window: 1m
      granularity: 1s
EOF

The requestRate metric scales based on requests per second, averaged over the configured window. A targetValue: 5 means the add-on targets 5 requests per second per replica. We use a low value here so that scaling is visible during testing. See Scaling for details on scaling metrics and how to tune them.

We verify the InterceptorRoute is ready:

kubectl get interceptorroute -n demo

We will see:

NAME         TARGETSERVICE   READY   AGE
sample-app   sample-app      True    10s

Step 5: Create a ScaledObject

The ScaledObject tells KEDA how to scale our sample-app deployment. It uses the external-push trigger type, which receives metrics from the HTTP Add-on’s scaler component.

kubectl apply -f - <<EOF
apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: sample-app
  namespace: demo
spec:
  scaleTargetRef:
    name: sample-app
  minReplicaCount: 0
  maxReplicaCount: 10
  cooldownPeriod: 30
  triggers:
    - type: external-push
      metadata:
        scalerAddress: keda-add-ons-http-external-scaler.keda:9090
        interceptorRoute: sample-app
EOF

The interceptorRoute value must match the name of the InterceptorRoute we created in the previous step. See Architecture for details on how these components connect.

We verify the ScaledObject was created:

kubectl get scaledobject -n demo

We will see:

NAME         SCALETARGETKIND      SCALETARGETNAME   MIN   MAX   TRIGGERS        ...
sample-app   apps/v1.Deployment   sample-app        0     10    external-push   ...

Step 6: Send Traffic and Observe Scaling

Now we test that autoscaling works. Since there is no traffic, KEDA has scaled the deployment to 0 replicas. We verify this:

kubectl get deployment sample-app -n demo

NAME         READY   UP-TO-DATE   AVAILABLE   AGE
sample-app   0/0     0            0           2m

For testing, we use kubectl port-forward to access the interceptor proxy. In production, your ingress or gateway must route traffic to the interceptor proxy service (keda-add-ons-http-interceptor-proxy) instead of directly to your application — see Configure Ingress for details.

kubectl port-forward -n keda svc/keda-add-ons-http-interceptor-proxy 8090:8080

In another terminal, we send a request with the matching Host header:

curl -H "Host: sample-app.example.com" localhost:8090

The first request may take a few seconds. This is the cold start: KEDA is scaling the deployment from 0 to 1 replica, and the interceptor holds the request until the pod is ready. We will see a response from the sample app once the pod starts.

We check replicas again:

kubectl get deployment sample-app -n demo

We will see 1 replica running:

NAME         READY   UP-TO-DATE   AVAILABLE   AGE
sample-app   1/1     1            1           3m

Step 7: Generate Load and Watch It Scale Up

To see scaling beyond 1 replica, we generate a burst of traffic. The wait=50ms query parameter tells whoami to hold each response for 50 milliseconds, which produces a steady rate of about 20 requests per second — enough to trigger scaling with our targetValue of 5:

for i in $(seq 1 300); do curl -s -H "Host: sample-app.example.com" "localhost:8090/?wait=50ms" > /dev/null; done

After the burst finishes, we check the deployment:

kubectl get deployment sample-app -n demo

We will see the replica count has increased:

NAME         READY   UP-TO-DATE   AVAILABLE   AGE
sample-app   2/2     2            2           5m

Step 8: Observe Scale to Zero

After the burst ends and the cooldown period passes (30 seconds, as configured in our ScaledObject), KEDA scales the deployment back to 0. We can watch this happen:

kubectl get deployment sample-app -n demo -w

We will see replicas decrease to 0:

NAME         READY   UP-TO-DATE   AVAILABLE   AGE
sample-app   2/2     2            2           5m
sample-app   1/1     1            1           6m
sample-app   0/0     0            0           7m

Step 9: Clean Up

To remove the sample application and all its resources:

kubectl delete namespace demo

What’s Next

Getting Help