Autoscaling a Dapr app with KEDA

Dapr is a programming model that’s being installed and operated using a sidecar, and thus leaves autoscaling to the hosting layer, for example Kubernetes. Many of Dapr’s bindings overlap with those of KEDA, an Event Driven Autoscaler for Kubernetes.

For apps that use these bindings, it is easy to configure a KEDA autoscaler.

Install KEDA

To install KEDA, follow these instructions on the KEDA Github page.

Create KEDA enabled Dapr binding

For this example, we’ll be using Kafka.
You can install Kafka in your cluster by using Helm:

$ helm repo add incubator http://storage.googleapis.com/kubernetes-charts-incubator
$ helm install --name my-kafka incubator/kafka

Next, we’ll create the Dapr Kafka binding for Kubernetes.
Paste the following in a file named kafka.yaml:

apiVersion: dapr.io/v1alpha1
kind: Component
metadata:
  name: kafkaevent
spec:
  type: bindings.kafka
  metadata:
  - name: brokers
    value: "my-kafka:9092"
  - name: topics
    value: "myTopic"
  - name: consumerGroup
    value: "group1"

The following YAML defines a Kafka component that listens for the topic myTopic, with consumer group group1 and that connects to a broker at my-kafka:9092.

Deploy the binding to the cluster:

$ kubectl apply -f kafka.yaml

Create the KEDA autoscaler for Kafka

Paste the following to a file named kafka_scaler.yaml, and put the name of your Deployment in the required places:

apiVersion: keda.k8s.io/v1alpha1
kind: ScaledObject
metadata:
  name: kafka-scaler
  namespace: default
  labels:
    deploymentName: <REPLACE-WITH-DEPLOYMENT-NAME>
spec:
  scaleTargetRef:
    deploymentName: <REPLACE-WITH-DEPLOYMENT-NAME>
  triggers:
  - type: kafka
    metadata:
      type: kafkaTrigger
      direction: in
      name: event
      topic: myTopic
      brokers:  my-kafka:9092
      consumerGroup: group2
      dataType: binary
      lagThreshold: '5'

Deploy the KEDA scaler to Kubernetes:

$ kubectl apply -f kafka_scaler.yaml

All done!

You can now start publishing messages to your Kafka topic myTopic and watch the pods autoscale when the lag threshold is bigger than 5, as defined in the KEDA scaler manifest.