message-queue-integrations

Message Queue Integrations Example

With this example, we demonstrate how to use each of the message queues supported within llama-deploy. Specifically, there is only one single example application, for which we show how to use the various message queues.

Setup

Prerequisites: Must have docker installed. (See here for how to install Docker Desktop which comes with docker-compose.)

Before we can deploy the app, we need to first install it. To do so, we use poetry:

cd _app
poetry shell
poetry install
cd ../

The above command builds the application and creates the poetry.lock file which is required in building the Docker image.

Finally, we also need to provide an OPENAI_API_KEY as the workflow in the app requires it. We provide this key in the template.env.openai file. After supplying the key, rename the file to .env.openai (i.e., drop "template" from the file name). Our Docker orchestration looks for this .env.openai file so make sure not to miss this step.

Deploying the app

Now, to deploy the app we need a single docker compose command. (All of the message queues have their own docker-compose.yaml file contained in their respective subfolders.)

Using any one of the commands (while in the parent /message-queue-integrations folder) below will spin the example app using the associated message queue.

# simple message queue
docker compose -f ./simple/docker/docker-compose.yml --project-directory ./ up --build -d

# aws
docker compose -f ./aws/docker/docker-compose.yml --project-directory ./ up --build -d

# kafka
docker compose -f ./kafka/docker/docker-compose.yml --project-directory ./ up --build -d

# rabbitmq
docker compose -f ./rabbitmq/docker/docker-compose.yml --project-directory ./ up --build -d

# redis
docker compose -f ./redis/docker/docker-compose.yml --project-directory ./ up --build -d

# solace
docker compose -f ./solace/docker/docker-compose.yml --project-directory ./ up --build -d

NOTE: In a real-world app, you would only use one of these message queues. So, in building your application (i.e., see _app/) you would not need to include the other message queues in your source code as we've done here.

Interacting with the system

Once the system is up and running, we can interact with it using the LlamaDeployClient:

from llama_deploy import LlamaDeployClient
from llama_deploy.control_plane.server import ControlPlaneConfig

# Set up control plane configuration
control_plane_config = ControlPlaneConfig(host="0.0.0.0", port=8000)

# Create a client to interact with the deployed system
client = LlamaDeployClient(control_plane_config)

# Start a session and run the funny joke workflow
session = client.create_session()
result = session.run("funny_joke_workflow", input="llamas")

# Print the result
print(result)

Deploying with Kubernetes

Some of the message queue folders contain a /kubernetes subfolder, where Kubernetes manifests that are used to deploy the example app with K8s can be found. Refer to the README contained in the /kubernetes folder for instructions on how to use those manifests to deploy the app with Kubernetes.