Using a Custom Worker Dockerfile for a Prediction Microservice

Custom Dockerfile for Prediction Worker

For any models that need specific libraries features that cannot be installed through pip or require a custom Dockerfile:

Step 1

In combinedtechstack/.env file add variable

PREDICTION_YOUR_MODEL=Your_model_folder_name

where Your_model_folder_name is the folder name in combinedtechstack/prediction/models containing the model and PREDICTION_YOUR_MODEL is the name of your model that will be referenced later (such as PREDICTION_SCENE_DETECT).

Step 2

Within the combinedtechstack/prediction/models/your_model_here add your custom Dockerfile.

Step 3

Navigate to the COMBINEDTECHSTACK/docker-compose.yml file and within it find the Prediction Microservice Workers section.

Step 4

Create a new worker in the docker-compose.yml:

worker_your_model:  # In this line, change the "your_model" to a better name  
    container_name: ${PREDICTION_YOUR_MODEL}_worker
    command: python3 worker.py
    build:
      context: .
      dockerfile: prediction/models/${PREDICTION_YOUR_MODEL}/Dockerfile
      args:
        - MODEL_NAME=${PREDICTION_YOUR_MODEL}
    volumes:
      - ./prediction/models/${PREDICTION_YOUR_MODEL}:/app/model
      - prediction_images:/app/images
    environment:
      - GUNICORN_CMD_ARGS=--reload
      - API_KEY=${API_KEY_PREDICTION}
      - SERVER_SOCKET=${SERVER_SOCKET}
    depends_on:
      - redis
      - server

Where PREDICTION_YOUR_MODEL will be automatically filled by Docker with the model paramater given in Step 1. Notably dockerfile: prediction/models/${PREDICTION_YOUR_MODEL}/Dockerfile will point to your custom Dockerfile. The worker will now be built using the custom Dockerfile.

For an example of a working Dockerfile in the model, see the: Speech Rec NER Microservice Custom Dockerfile

(Documentation by Christopher Doan)