Edit

Share via

Scale Azure OpenAI for Python chat using RAG with Azure Container Apps

  • Article

Learn how to add load balancing to your application to extend the chat app beyond the Azure OpenAI token and model quota limits. This approach uses Azure Container Apps to create three Azure OpenAI endpoints, as well as a primary container to direct incoming traffic to one of the three endpoints.

This article requires you to deploy two separate samples:

  • Chat app

    • If you haven't deployed the chat app yet, wait until after the load balancer sample is deployed.

    • If you have already deployed the chat app once, you'll change the environment variable to support a custom endpoint for the load balancer and redeploy it again.

    • Chat app available in these languages:

  • Load balancer app

Note

This article uses one or more AI app templates as the basis for the examples and guidance in the article. AI app templates provide you with well-maintained, easy to deploy reference implementations that help to ensure a high-quality starting point for your AI apps.

Architecture for load balancing Azure OpenAI with Azure Container Apps

Because the Azure OpenAI resource has specific token and model quota limits, a chat app using a single Azure OpenAI resource is prone to have conversation failures due to those limits.

Diagram showing chat app architecture with Azure OpenAI resource highlighted.

To use the chat app without hitting those limits, use a load balanced solution with Azure Container Apps. This solution seamlessly exposes a single endpoint from Azure Container Apps to your chat app server.

Diagram showing chat app architecture with Azure Container Apps in front of three Azure OpenAI resources.

The Azure Container app sits in front of a set of Azure OpenAI resources. The Container app solves two scenarios: normal and throttled. During a normal scenario where token and model quota is available, the Azure OpenAI resource returns a 200 back through the Container App and App Server.

Diagram displaying a normal scenario. The normal scenario shows three Azure OpenAI endpoint groups with the first group of two endpoints getting successful traffic.

When a resource is in a throttled scenario such as due to quota limits, the Azure Container app can retry a different Azure OpenAI resource immediately to fullful the original chat app request.

Diagram displaying a throttling scenario with a 429 failing response code and a response header of how many seconds the client has to wait to retry.

Prerequisites

Open Container apps local balancer sample app

GitHub Codespaces runs a development container managed by GitHub with Visual Studio Code for the Web as the user interface. For the most straightforward development environment, use GitHub Codespaces so that you have the correct developer tools and dependencies preinstalled to complete this article.

Open this project in GitHub Codespaces

Important

All GitHub accounts can use Codespaces for up to 60 hours free each month with 2 core instances. For more information, see GitHub Codespaces monthly included storage and core hours.

Deploy Azure Container Apps load balancer

  1. Sign in to the Azure Developer CLI to provide authentication to the provisioning and deployment steps.

    azd auth login --use-device-code

  2. Set an environment variable to use Azure CLI authentication to the post provision step.

    azd config set auth.useAzCliAuth "true"

  3. Deploy the load balancer app.

    azd up

    You'll need to select a subscription and region for the deployment. These don't have to be the same subscription and region as the chat app.

  4. Wait for the deployment to complete before continuing.

Get the deployment endpoint

  1. Use the following command to display the deployed endpoint for the Azure Container app.

    azd env get-values

  2. Copy the CONTAINER_APP_URL value. You will use it in the next section.

Redeploy Chat app with load balancer endpoint

These are completed on the chat app sample.

  1. Open the chat app sample's dev container using one of the following choices.

    Language Codespaces Visual Studio Code
    .NET Open in GitHub Codespaces Open in Dev Containers
    JavaScript Open in GitHub Codespaces Open in Dev Containers
    Python Open in GitHub Codespaces Open in Dev Containers

  2. Sign in to Azure Developer CLI (AZD).

    azd auth login

    Finish the sign in instructions.

  3. Create an AZD environment with a name such as chat-app.

    azd env new <name>

  4. Add the following environment variable, which tells the Chat app's backend to use a custom URL for the OpenAI requests.

    azd env set OPENAI_HOST azure_custom

  5. Add the following environment variable, substituting <CONTAINER_APP_URL> for the URL from the previous section. This action tells the Chat app's backend what the value is of the custom URL for the OpenAI request.

    azd env set AZURE_OPENAI_CUSTOM_URL <CONTAINER_APP_URL>

  6. Deploy the chat app.

    azd up

You can now use the chat app with the confidence that it's built to scale across many users without running out of quota.

Stream logs to see the load balancer results

  1. In the Azure portal, search your resource group.

  2. From the list of resources in the group, select the Container App resource.

  3. Select Monitoring -> Log stream to view the log.

  4. Use the chat app to generate traffic in the log.

  5. Look for the logs, which reference the Azure OpenAI resources. Each of the three resources has its numeric identity in the log comment beginning with Proxying to https://openai3, where 3 indicates the third Azure OpenAI resource.

    Screenshot showing Azure Container App streaming logs with two log lines high lighted which demonstrate the log comment.

  6. As you use the chat app, when the load balancer receives status that the request has exceeded quota, the load balancer automatically rotates to another resource.

Configure the tokens per minute quota (TPM)

By default, each of the OpenAI instances in the load balancer will be deployed with 30,000 TPM (tokens per minute) capacity. You can use the chat app with the confidence that it's built to scale across many users without running out of quota. Change this value when:

  • You get deployment capacity errors: lower that value.
  • Planning higher capacity, raise the value.

  1. Use the following command to change the value.

    azd env set OPENAI_CAPACITY 50

  2. Redeploy the load balancer.

    azd up

Clean up resources

When you're done with both the chat app and the load balancer, clean up the resources. The Azure resources created in this article are billed to your Azure subscription. If you don't expect to need these resources in the future, delete them to avoid incurring more charges.

Clean up chat app resources

Return to the chat app article to clean up those resources.

Clean upload balancer resources

Run the following Azure Developer CLI command to delete the Azure resources and remove the source code:

azd down --purge --force

The switches provide:

  • purge: Deleted resources are immediately purged. This allows you to reuse the Azure OpenAI TPM.
  • force: The deletion happens silently, without requiring user consent.

Clean up GitHub Codespaces

Deleting the GitHub Codespaces environment ensures that you can maximize the amount of free per-core hours entitlement you get for your account.

Important

For more information about your GitHub account's entitlements, see GitHub Codespaces monthly included storage and core hours.

  1. Sign into the GitHub Codespaces dashboard (https://github.com/codespaces).

  2. Locate your currently running Codespaces sourced from the azure-samples/openai-aca-lb GitHub repository.

    Screenshot of all the running Codespaces including their status and templates.

  3. Open the context menu for the codespace and then select Delete.

    Screenshot of the context menu for a single codespace with the delete option highlighted.

Get help

If you have trouble deploying the Azure API Management load balancer, log your issue to the repository's Issues.

Sample code

Samples used in this article include:

Next step