compose.yaml

# This docker compose file builds or pulls the images and runs the containers of all services on our local computer.
# Docker Compose automatically creates a user-defined bridge network for the containers to communicate with each other.
# The only part of the app we can reach is the nginx container, whose port 80 we publish to our host machine's port 80.
# The app wll be available at localhost.

# To build and run all services:
#  `docker-compose up --build`
# To tear down the app with all its services
#  `docker compose down`

services:

  # Build UI image and run container once the db-api and lm-api are up. Pass their hostnames as env vars.
  chat-ui:
    build: ./chat-ui
    depends_on:
      db-api:
        condition: service_started
      lm-api:
        condition: service_started
    environment:
      DB_API_URL: http://db-api
      LM_API_URL: http://lm-api
      SESSION_KEY: top-secret-key

  # Build db-api image and run container once the db is up. Pass db credentials and hostname as env vars.
  db-api:
    build: ./db-api
    depends_on:
      db:
        condition: service_healthy
    environment:
        POSTGRES_USER: myuser
        POSTGRES_PASSWORD: mypassword
        POSTGRES_DB: postgres
        POSTGRES_HOST: db

  # Pull postgres image and run container. Set db credentials as env vars and perform periodical health check.
  db:
    image: postgres
    restart: always
    environment:
      POSTGRES_USER: myuser  # set database username
      POSTGRES_PASSWORD: mypassword # set database password
    healthcheck: # healthcheck to ensure the database is up and running before the db-api container starts
      test: [ "CMD-SHELL", "pg_isready -U myuser -h localhost -d postgres" ]
      interval: 1s
      timeout: 5s
      retries: 10

  # Pull llama.cpp server image and run container; pass parameters to llama.cpp process; mount model file directory.
  lm-api:
    image: ghcr.io/ggerganov/llama.cpp:server
    volumes:
      - ./lm-api/models:/models
    # Run local inference server with max 200 tokens and flash attention
    command: -m /models/qwen2.5-0.5b-instruct-q5_k_m.gguf --port 80 --host 0.0.0.0 -n 128 -fa -c 1024

  # Pull nginx image and run container once chat-ui is up. Mount the nginx configuration file and publish port 80 to host machine.
  nginx:
    image: nginx:latest
    volumes:
      - ./nginx/nginx.conf:/etc/nginx/nginx.conf
    ports:
      - "80:80"
    depends_on:
      - chat-ui