Version: 3.x

Deploying Your Rasa Assistant

This page explains when and how to deploy an assistant built with Rasa. It will allow you to make your assistant available to users and set you up with a production-ready environment.

When to Deploy Your Assistant

The best time to deploy your assistant and make it available to test users is once it can handle the most important happy paths or is what we call a minimum viable assistant.

The recommended deployment methods described below make it easy to share your assistant with test users via the share your assistant feature in Rasa X. Then, when you're ready to make your assistant available via one or more Messaging and Voice Channels, you can easily add them to your existing deployment set up.

Recommended Deployment Methods

The recommended way to deploy an assistant is using either the Server Quick-Install or Helm Chart options we support. Both deploy Rasa X and your assistant. They are the easiest ways to deploy your assistant, allow you to use Rasa X to view conversations and turn them into training data, and are production-ready. For more details on deployment methods see the Rasa X Installation Guide.

Server Quick-Install

The Server Quick-Install script is the easiest way to deploy Rasa X and your assistant. It installs a Kubernetes cluster on your machine with sensible defaults, getting you up and running in one command.

Helm Chart

For assistants that will receive a lot of user traffic, setting up a Kubernetes or Openshift deployment via our Helm charts is the best option. This provides a scalable architecture that is also straightforward to deploy. However, you can also customize the Helm charts if you have specific requirements.

Deploying a Rasa Open Source Assistant

While the above deployment methods involve deploying an assistant with Rasa X, the following instructions describe how to deploy a Rasa Open Source server only by using the Rasa Helm Chart in a scalable cluster environment using OpenShift or Kubernetes (K8S).

Cluster Requirements

To install the Rasa Helm chart, you need an existing Kubernetes cluster or OpenShift cluster. Setting up a Kubernetes / OpenShift cluster can be tedious, hence we recommend to get a managed cluster from a cloud provider like Google Cloud, DigitalOcean, Microsoft Azure, or Amazon EKS.

note

The Rasa Helm chart is open source and available in the helm-charts repository. Please create an issue in this repository if you discover bugs or have suggestions for improvements.

Installation Requirements

  1. Check that you have installed the Kubernetes or OpenShift command line interface (CLI). You can check this using the following command:

    kubectl version --short --client
    # The output should be similar to this
    # Client Version: v1.19.11

    If this command resulted in an error, please install the Kubernetes CLI or the OpenShift CLI depending on the cluster you’re using.

  2. Make sure that the Kubernetes / OpenShift CLI is correctly connected to your cluster. You can do so by using the following commands:

    kubectl version --short
    # The output should be similar to this
    # Client Version: v1.19.11
    # Server Version: v1.19.10

    If you get an error when executing the command, you are not connected to your cluster. To get the command to connect to the cluster please consult your cluster’s admin or the documentation of your cloud provider.

  3. Make sure you have the Helm CLI installed. To check this, run:

    helm version --short
    # The output should be similar to this
    # v3.6.0+g7f2df64

    If this command leads to an error, please install the Helm CLI.

    In case you are using a version <3.5 of Helm, please update to Helm version >=3.5.

Installation

1. Create Namespace

We recommend installing Rasa Open Source in a separate namespace to avoid interfering with existing cluster deployments. To create a new namespace run the following command:

kubectl create namespace <your namespace>

2. Create Values File

Prepare an empty file called rasa-values.yml which will include all your custom configuration for the installation with Helm.

All available values you can find in the Rasa helm chart repository.

note

The default configuration of the Rasa chart deploys a Rasa Open Source Server, downloads a model, and serves the downloaded model. Visit the Rasa helm chart repository to check out more examples of configuration.

3. Loading an initial model

The first time you install Rasa, you may not have a model server available yet, or you may want an lightweight model for testing the deployment. For this purpose, you can choose between training or downloading an initial model. By default, the Rasa chart downloads an example model from GitHub. To use this option, you don't have to change anything.

If you want to define an existing model to download from a URL you define instead, update your rasa-values.yaml with the URL according to the following configuration:

applicationSettings:
initialModel: "https://github.com/RasaHQ/rasa-x-demo/blob/master/models/model.tar.gz?raw=true"
note

The URL for the initial model download has to point to a tar.gz file and must not require authentication.

If you want to train an initial model you can do this by setting the applicationSettings.trainInitialModel to true. It creates a init container that trains a model based on data located in the /app directory. If the /app directory is empty it creates a new project. You can find an example that shows how to download data files from a git repository and train an initial model in the Rasa helm charts examples.

4. Deploy Rasa Open Source Assistant

Run the following commands:

# Add the repository which contains the Rasa Helm chart
helm repo add rasa https://helm.rasa.com
# Deploy Rasa Open Source
helm install \
--namespace <your namespace> \
--values rasa-values.yml \
<release name> \
rasa/rasa
note

OpenShift only: If the deployment fails and oc get events returns 1001 is not an allowed group spec.containers[0].securityContext.securityContext.runAsUser, re-run the installation command with the following values:

postgresql:
volumePermissions:
securityContext:
runAsUser: "auto"
securityContext:
enabled: false
shmVolume:
chmod:
enabled: false
nginx:
image:
name: nginxinc/nginx-unprivileged
port: 8080

Then wait until the deployment is ready. If you want to check on its status, the following command will block until the Rasa deployment is ready:

kubectl --namespace <your namespace> \
wait \
--for=condition=available \
--timeout=20m \
--selector app.kubernetes.io/instance=<release name> \
deployment

5. Access Rasa Open Source Assistant

By default the Rasa deployment is exposed via the rasa (<release name>) service and accessible only within a Kubernetes cluster. You can get the IP address using this command:

export SERVICE_PORT=$(kubectl get --namespace <your namespace> -o jsonpath="{.spec.ports[0].port}" services <release name>)
kubectl port-forward --namespace <your namespace> svc/<release name> ${SERVICE_PORT}:${SERVICE_PORT} &

You can then access the deployment on http://127.0.0.1:${SERVICE_PORT}

Visit the Rasa helm chart README to learn other ways to expose your deployment.

Next Steps

Visit the Rasa helm chart repository where you can find examples of configuration and learn how to e.g. integrate your Rasa Open Source deployment with Rasa X.

Alternative Deployment Methods

Docker Compose

You can also run Rasa X in a Docker Compose setup, without the cluster environment. We have an install script for doing so, as well as manual instructions for any custom setups.

Rasa Open Source Only Deployment

It is also possible to deploy a Rasa assistant without Rasa X using Docker Compose. To do so, you can build your Rasa Assistant locally or in Docker. Then you can deploy your model in Docker Compose.

Deploying Your Action Server

Building an Action Server Image

If you build an image that includes your action code and store it in a container registry, you can run it as part of your deployment, without having to move code between servers. In addition, you can add any additional dependencies of systems or Python libraries that are part of your action code but not included in the base rasa/rasa-sdk image.

Automating your Action Server Image Builds

In addition to a manually creating a new Action Server image, you can use the Rasa Action Server GitHub Action to automate image builds. If GitHub Actions are new to you, it might be helpful to get familiar with GitHub Actions Documentation.

The following steps assume that you already created a GitHub repository and you have a DockerHub account.

To create a workflow for building and pushing a Docker image into a DockerHub registry:

  1. Add GitHub Secrets with your DockerHub login name and password. You can find details on how to create encrypted secrets for a repository in the Github docs

    The example uses the following secrets:

    • DOCKER_HUB_LOGIN - a login name for DockerHub
    • DOCKER_HUB_PASSWORD - a password for DockerHub

  2. In your GitHub repository create a file .github/workflows/action_server.yml.

The GitHub Action workflow below builds a new docker image every time files inside the actions/ directory have changed and the changes are pushed into the main branch.

on:
push:
branches:
- main
paths:
- 'actions/**'
jobs:
build_and_deploy:
runs-on: ubuntu-latest
name: Build Action Server image and upgrade Rasa X deployment
steps:
- name: Checkout repository
uses: actions/checkout@v2
- id: action_server
name: Build an action server with a custom actions
uses: RasaHQ/action-server-gha@master
# Full list of parameters: https://github.com/RasaHQ/action-server-gha/tree/master#input-arguments
with:
docker_image_name: 'account_username/repository_name'
docker_registry_login: ${{ secrets.DOCKER_HUB_LOGIN }}
docker_registry_password: ${{ secrets.DOCKER_HUB_PASSWORD }}
# More details about github context:
# https://docs.github.com/en/actions/reference/context-and-expression-syntax-for-github-actions#github-context
#
# github.sha - The commit SHA that triggered the workflow run
docker_image_tag: ${{ github.sha }}

  1. Push your changes to the main branch. After changes are pushed, the workflow will build and push a new image into the DockerHub registry.

  2. Now, you can use your new brand docker image.

  3. You can also extend your workflow, so that you do not have to manually update your Rasa X deployment. The example below shows how to extend your workflow with an additional step that updates a Rasa X Helm Chart deployment.

on:
push:
branches:
- main
jobs:
build_and_deploy:
runs-on: ubuntu-latest
name: Build Action Server image and upgrade Rasa X deployment
steps:
[..]
# This step shows only the example of output parameter usage
# and it's not focused on deployment itself.
- name: "Upgrade a Rasa X deployment"
run: |
helm upgrade --install --reuse-values \
--set app.name=${{ steps.action_server.outputs.docker_image_name }} \
--set app.tag=${{ steps.action_server.outputs.docker_image_tag }} rasa rasa-x/rasa-x

As you can see it's possible to use output variables from the action_server step. The steps.action_server.outputs.docker_image_name variable returns a docker image name and the steps.action_server.outputs.docker_image_tag variable returns a docker image tag.

More examples on how to use and customize Rasa GitHub Actions you can find in the Rasa GitHub Actions repository.

Manually Building an Action Server

To create your image:

  1. Make sure your actions are defined in actions/actions.py. The rasa/rasa-sdk image will automatically look for the actions in this file.

  2. If your actions have any extra dependencies, create a list of them in a file, actions/requirements-actions.txt.

  3. Create a file named Dockerfile in your project directory, in which you'll extend the official SDK image, copy over your code, and add any custom dependencies (if necessary). For example:

    # Extend the official Rasa SDK image
    FROM rasa/rasa-sdk:3.1.1
    # Use subdirectory as working directory
    WORKDIR /app
    # Copy any additional custom requirements, if necessary (uncomment next line)
    # COPY actions/requirements-actions.txt ./
    # Change back to root user to install dependencies
    USER root
    # Install extra requirements for actions code, if necessary (uncomment next line)
    # RUN pip install -r requirements-actions.txt
    # Copy actions folder to working directory
    COPY ./actions /app/actions
    # By best practices, don't run the code with root user
    USER 1001

You can then build the image via the following command:

docker build . -t <account_username>/<repository_name>:<custom_image_tag>

The <custom_image_tag> should reference how this image will be different from others. For example, you could version or date your tags, as well as create different tags that have different code for production and development servers. You should create a new tag any time you update your code and want to re-deploy it.

Using your Custom Action Server Image

If you're building this image to make it available from another server, for example a Rasa X or Rasa Enterprise deployment, you should push the image to a cloud repository.

This documentation assumes you are pushing your images to DockerHub. DockerHub will let you host multiple public repositories and one private repository for free. Be sure to first create an account and create a repository to store your images. You could also push images to a different Docker registry, such as Google Container Registry, Amazon Elastic Container Registry, or Azure Container Registry.

You can push the image to DockerHub via:

docker login --username <account_username> --password <account_password>
docker push <account_username>/<repository_name>:<custom_image_tag>

To authenticate and push images to a different container registry, please refer to the documentation of your chosen container registry.

How you reference the custom action image will depend on your deployment. Pick the relevant documentation for your deployment: