docker-compose.yml

# By default, we disable mountain the current directory under /modyn_host. However, this might be helpful for local development.
# For the trainer server, you additionally might want to enable the runtime and deployment option to enable the GPU in the container.
# For the storage, you probably want to mount some volume containing the datasets.
# For the selector, you might want to mount a volume on a fast local storage device to store the trigger training sets.
# For storage-db and metadata-db, you might want to increase the amount of available /dev/shm and store the postgres files on a fast local disk.

services:
  metadata-db:
    image: postgres:16.4-alpine
    restart: always
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
      POSTGRES_DB: postgres
      POSTGRES_HOST_AUTH_METHOD: md5
    command: postgres -c config_file=/etc/postgresql/postgresql.conf
    #    shm_size: 24gb
    volumes:
      - ./metadata-postgres-data:/var/lib/postgresql/data
      - ./conf/metadata_postgresql.conf:/etc/postgresql/postgresql.conf
      - ./conf/pg_hba.conf:/tmp/pg_hba.conf
      - ./conf/init_pg_hba.sh:/docker-entrypoint-initdb.d/init_pg_hba.sh
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 5s
      timeout: 5s
      retries: 20
  storage-db:
    image: postgres:16.4-alpine
    restart: always
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
      POSTGRES_DB: postgres
      POSTGRES_HOST_AUTH_METHOD: md5
    command: postgres -c config_file=/etc/postgresql/postgresql.conf
    #    shm_size: 8gb
    volumes:
      - ./storage-postgres-data:/var/lib/postgresql/data
      - ./conf/storage_postgresql.conf:/etc/postgresql/postgresql.conf
      - ./conf/pg_hba.conf:/tmp/pg_hba.conf
      - ./conf/init_pg_hba.sh:/docker-entrypoint-initdb.d/init_pg_hba.sh
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 5s
      timeout: 5s
      retries: 20
  storage:
    restart: on-failure
    depends_on:
      storage-db:
        condition: service_healthy
    build:
      context: .
      dockerfile: docker/Storage/Dockerfile
    volumes:
      - storage-data:/app/storage
  #      - /mnt/datasets:/datasets
  #      - .:/modyn_host
  metadata_processor:
    restart: on-failure
    depends_on:
      metadata-db:
        condition: service_healthy
    build:
      context: .
      dockerfile: docker/MetadataProcessor/Dockerfile
  model_storage:
    restart: on-failure
    depends_on:
      metadata-db:
        condition: service_healthy
    build:
      context: .
      dockerfile: docker/Model_Storage/Dockerfile
    volumes:
      - model_storage-data:/tmp/models
  evaluator:
    restart: on-failure
    depends_on:
      - model_storage
      - storage
    build:
      context: .
      dockerfile: docker/Evaluator/Dockerfile
  trainer_server:
    restart: on-failure
    depends_on:
      - storage
      - selector
      - model_storage
      - metadata-db
    build:
      context: .
      dockerfile: docker/Trainer_Server/Dockerfile
    volumes:
      - downsampling-data:/tmp/offline_dataset # Remove when using a fast other storage directory (line below)
  #      - /mnt/ssd/offline_dataset:/tmp/offline_dataset
  #      - .:/modyn_host
  #    shm_size: 4gb
  # CUDASTART (ignore/delete if enabling cuda manually)
  #    runtime: nvidia
  #    deploy:
  #      resources:
  #        reservations:
  #          devices:
  #            - driver: nvidia
  #              count: 1
  #              capabilities: [gpu, utility, compute]
  # CUDAEND (ignore/delete if enabling cuda manually)
  selector:
    restart: on-failure
    depends_on:
      metadata-db:
        condition: service_healthy
    build:
      context: .
      dockerfile: docker/Selector/Dockerfile
    volumes:
      - selector-localstorage-data:/tmp/local_storage # Remove when using a fast other storage directory (lines below)
      - selector-data:/tmp/trigger_samples # Remove when using a fast other storage directory (lines below)
  #      - /mnt/ssd/local_storage:/tmp/local_storage
  #      - /mnt/ssd/trigger_samples:/tmp/trigger_samples
  #      - .:/modyn_host
  #    shm_size: 4gb
  supervisor:
    restart: on-failure
    depends_on:
      - storage
      - metadata-db
      - storage-db
      - selector
      - metadata_processor
      - trainer_server
      - evaluator
    build:
      context: .
      dockerfile: docker/Supervisor/Dockerfile
    volumes:
      - supervisor-evaluation-data:/tmp/evaluation_results
    #      - .:/modyn_host
    container_name: supervisor
    ports:
      - "3000:50063"
  tests:
    depends_on:
      supervisor:
        condition: service_started
      storage:
        condition: service_started
      selector:
        condition: service_started
      metadata_processor:
        condition: service_started
      model_storage:
        condition: service_started
      evaluator:
        condition: service_started
      trainer_server:
        condition: service_started
      metadata-db:
        condition: service_healthy
      storage-db:
        condition: service_healthy
    build:
      context: .
      dockerfile: docker/Tests/Dockerfile
    volumes:
      - storage-data:/app/storage
volumes:
  storage-data:
  selector-data:
  selector-localstorage-data:
  downsampling-data:
  model_storage-data:
  supervisor-evaluation-data: