Skip to main content

Deploying to Kubernetes

Kubernetes (and OpenShift) provide a reliable way to run containerized applications at scale. With Kubernetes, you can:

  • Orchestrate multiple Rasa Pro services (e.g., Rasa core container, Action Server, Rasa Pro Services) on any cloud or on-prem setup.
  • Easily scale up or down by adding more replicas.
  • Seamlessly manage rolling updates, networking, and load balancing.
  • Simplify the deployment of new versions of your assistant.

If you are unfamiliar with Kubernetes or want a fully managed solution, consider Rasa’s Managed Service.

Deployment Requirements​

Before deploying Rasa Pro on Kubernetes, make sure you have:

  1. A Kubernetes or OpenShift cluster
    • Many providers (AWS EKS, Azure AKS, GCP, DigitalOcean) offer managed clusters.
    • Ensure you have kubectl (for Kubernetes) or oc (for OpenShift) installed and connected to your cluster.
  2. Helm CLI (v3.5 or newer)
    • You’ll need it to install the Rasa Pro Helm chart.
  3. A valid Rasa Pro License
    • You will pass it as a secret or an environment variable in your deployment.
  4. (Optional) A Model Storage Bucket
    • If you plan to store or mount your trained models from cloud storage (AWS S3, GCP Storage, or Azure Blob), set this up in advance.
  5. (Optional) A Kafka cluster and Data Warehouse
    • Required if you plan to deploy Rasa Pro Services for analytics and logging.

If you do not already have the above, see your cloud provider’s documentation for setting up a Kubernetes or OpenShift cluster. For additional details on Rasa Pro environment variables or advanced configuration, refer to the Reference.

How to Deploy Rasa Pro​

1. Kubernetes/OpenShift Cluster​

  • Confirm connectivity:

    kubectl version

    Ensure it shows both client and server versions (for OpenShift, use oc version).

  • Create a dedicated namespace (recommended):

    kubectl create namespace <your-namespace>
    kubectl config set-context --current --namespace=<your-namespace>

    This helps isolate your Rasa Pro deployment from other workloads.

2. Rasa Pro Helm Chart​

Rasa Pro provides a Helm chart to simplify deployment. The chart is hosted on a public Artifact Registry.

  1. Download the Helm chart:

    helm pull oci://europe-west3-docker.pkg.dev/rasa-releases/helm-charts/rasa

    This command downloads a file named rasa-<version>.tgz.

  2. Check your Helm version:

    helm version --short

    You need v3.5 or newer.

3. Deploy Rasa Pro​

Below is the minimal workflow for deploying Rasa Pro on Kubernetes or OpenShift using the Helm chart.

a) Prepare Secrets​

  1. Create a secrets.yml file (or name it as you wish) with the Rasa Pro license and any other secret values you may need (authentication tokens, etc.). Base64-encode your secret values.

    secrets.yml
    apiVersion: v1
    kind: Secret
    metadata:
      name: rasa-secrets
    type: Opaque
    data:
      rasaProLicense: <BASE64ENCODED_LICENSE>
      authToken: <BASE64ENCODED_VALUE>
      jwtSecret: <BASE64ENCODED_VALUE>

  2. Apply the secrets:

    kubectl apply -f secrets.yml

b) Create a values.yml for your deployment​

  1. Minimal values.yml example:

    values.yml
    # Rasa Pro Container
    rasa:
      image:
        repository: "europe-west3-docker.pkg.dev/rasa-releases/rasa-pro/rasa-pro"
        tag: "3.8.0-latest"
      # Additional Rasa configuration can go here.

    # Disable the Rasa Pro Services container if not needed
    rasaProServices:
      enabled: false

  2. Deploy with Helm:

    helm install \
        --namespace <your-namespace> \
        --values values.yml \
        <release-name> \
        rasa-<version>.tgz

    This starts a Rasa Pro pod. If you need to update any configuration:

    helm upgrade \
        --namespace <your-namespace> \
        --values values.yml \
        <release-name> \
        rasa-<version>.tgz

    To remove:

    helm delete <release-name>

4. Model Storage Bucket​

To load a trained model from cloud storage:

  1. Set up a bucket on AWS S3, Azure Blob, or Google Cloud Storage, and upload your trained model.
  2. Mount or configure the bucket for your Rasa container.

For example, in Google Cloud:

values.yml
rasa:
  endpoints:
    models:
      enabled: false

  volumes:
    - csi:
        driver: gcsfuse.csi.storage.gke.io
        readOnly: true
        volumeAttributes:
          bucketName: <YOUR BUCKET NAME>
          mountOptions: implicit-dirs,only-dir=<YOUR DIR>
      name: rasa-models

  volumeMounts:
    - name: rasa-models
      mountPath: /app/models
      readOnly: true

  serviceAccount:
    create: true
    annotations:
      iam.gke.io/gcp-service-account: <YOUR SERVICE ACCOUNT EMAIL>
    name: "rasa-pro-sa"

  podAnnotations:
    gke-gcsfuse/volumes: "true"

For other cloud platforms or more advanced configurations, see the Reference.

5. Deploy Action Server​

If your assistant uses Custom Actions, you can build and deploy a separate Action Server container alongside your Rasa Pro container.

  1. Build your custom action image:

    • Place your Python code in actions/actions.py.

    • Optionally specify any dependencies in requirements-actions.txt.

    • Create a Dockerfile extending the official rasa/rasa-sdk image:

      FROM rasa/rasa-sdk:latest
      WORKDIR /app

      # (Optional) for custom dependencies
      # COPY actions/requirements-actions.txt ./
      # RUN pip install -r requirements-actions.txt

      COPY ./actions /app/actions
      USER 1001

    • Build & push the image to your container registry (e.g., DockerHub, GCR, ECR).

  2. Reference your Action Server in values.yml:

    values.yml
    rasa:
      endpoints:
        actionEndpoint:
          url: http://action-server:5055/webhook

    actionServer:
      enabled: true
      image:
        repository: <my-docker-username>/<my-action-repo>
        tag: <custom-tag>

  3. Upgrade the Helm release:

    helm upgrade \
        --namespace <your-namespace> \
        --values values.yml \
        <release-name> \
        rasa-<version>.tgz

6. Deploy Rasa Pro Services​

Rasa Pro Services is an optional container providing analytics, data collection, and other enterprise features. It must connect to:

  • A Kafka cluster (production-ready)
  • A data warehouse (e.g., PostgreSQL)

  1. Configure Kafka for your Rasa Pro container. In your values.yml, make sure:

    values.yml
    rasa:
      settings:
        endpoints:
          eventBroker:
            enabled: true
            type: kafka
            # other settings as needed...

  2. Enable Rasa Pro Services:

    values.yml
    rasaProServices:
      enabled: true
      image:
        repository: "europe-west3-docker.pkg.dev/rasa-releases/rasa-pro/rasa-pro-services"
        tag: "3.2.0-latest"
      environmentVariables:
        KAFKA_BROKER_ADDRESS:
          value: "<KAFKA-BROKER-URL>"
        KAFKA_TOPIC:
          value: "rasa-assistant"
        RASA_ANALYTICS_DB_URL:
          secret:
            name: analytics-db-url
            key: dburl
        # More environment variables if needed...

  3. Upgrade via Helm:

    helm upgrade \
        --namespace <your-namespace> \
        --values values.yml \
        <release-name> \
        rasa-<version>.tgz

You can confirm the Rasa Pro Services pod is running and check the /healthcheck endpoint to verify status.

7. Adding Environment Variables​

You can pass extra environment variables to any container by adding them to values.yml. For example:

values.yml
rasa:
  extraEnvs:
    - name: "LOG_LEVEL"
      value: "warning"

If you have sensitive data (e.g., passwords), store them in Kubernetes secrets and reference them in values.yml. A simple example:

values.yml
rasa:
  extraEnvs:
    - name: "KAFKA_PASSWORD"
      valueFrom:
        secretKeyRef:
          name: kafka-secrets
          key: password

A full list of available environment variables for Rasa Pro, Rasa Pro Services, and the Rasa Action Server can be found in the Environment Variables reference.

Last updated on