Machine Learning Pipeline using FastAPI, RabbitMQ and image classifier (ONNX)

This inference pipeline is composed by a producer, a RabbitMQ queue and a consumer. FYI, in the text below each variable written in CAPITAL-YELLOW is defined in the docker-compose.yaml.

• The producer receives post requests (base64 encoded images) and pushes messages to the RabbitMQ queue where each message is composed by a unique id (UUID), an image, a CALLBACK_URL (webhook for asynchronous response) and the time.time() where the image was sent. If no response is received after TIMEOUT seconds, the client will receive an error 400. Check files at src/producer.
• The consumer receives simoultaneously up to MAX_BATCH_SIZE and processes them in batches. However, once the first message is received, all messages will be processed after BATCH_TIMEOUT seconds, even if they are fewer then MAX_BATCH_SIZE. Once a message is processed, it is removed from the RABBITMQ queue (channel.basic_ack) and is sent back to the producer.
• Response to client: once the processed message is sent back to the producer, if the response if received within a TIMEOUT time, then it is stored in a local dictionary. The initial asynchronous function used to receive client requests, checks every 0.01 seconds (sleeps asynchronously) if the server response is present within the local dictionary. If is present, it sends the response back to the client.

This pipeline could be further improved by using a "not TCP protocol" in the consumer-producer response.

How to run:

Build Images and run containers.

        docker compose up -d

FYI: RABBITMQ takes a while to setup, even if the container are present, wait some seconds before staring to send requests.

Test the service:

        bash send_requests.sh

How to configure rabbitMQ:

1. Create a password hash with:

       python src/rabbitmq/password.py your_password
   

2. Open src/rabbitmq/definitions.json, add create a dict within users:

   {
   "name": "your_user_name",
   "password_hash": "password_hash_created_at_point_1",
   "hashing_algorithm": "rabbit_password_hashing_sha256",
   "tags": ""
   }
   

3. Still in src/rabbitmq/definitions.json make sure the user has the the specified permission under the permissions key:

   {
   "user": "your_user_name",
   "vhost": "\/",
   "configure": ".*",
   "write": ".*",
   "read": ".*"
   }