Docker Compose-Spickzettel

Docker Compose describes a multi-container application in one YAML file and runs it with a single command. The file declares services (containers), plus the networks, volumes, secrets and configs they use.

ℹ️ The version: key at the top is obsolete — the Compose Spec ignores it. Start the file with services:.

📁 File Names & Precedence

# check what Compose actually resolved, after merges and interpolation
docker compose config

# use a specific set of files
docker compose -f compose.yaml -f compose.prod.yaml up -d

YAML rules that bite

• Two spaces per indent level, never tabs
• Keys and values are case-sensitive
• Lists use -; quote any string containing :, #, {, } or a leading *
• yes/no/on/off are booleans — quote them if you mean the words

🧱 Minimal Structure

services:
  web:
    image: nginx:1.27
    ports:
      - '8080:80'
    depends_on:
      - api

  api:
    build: .
    environment:
      DATABASE_URL: postgres://db:5432/app
    networks:
      - backend

networks:
  backend:

volumes:
  db-data:

⚙️ Service Options

services:
  app:
    # where the image comes from
    image: myapp:1.2
    build: .
    pull_policy: always

    # identity and lifecycle
    container_name: myapp
    hostname: app
    restart: unless-stopped
    init: true
    stop_grace_period: 30s

    # what it runs
    entrypoint: ['/entrypoint.sh']
    command: ['node', 'server.js']
    working_dir: /app
    user: '1000:1000'

    # configuration
    environment:
      NODE_ENV: production
      API_KEY: ${API_KEY}
    env_file:
      - .env
      - .env.production

    # connectivity
    ports:
      - '8080:80'
    expose:
      - '9000'
    networks:
      - frontend
    extra_hosts:
      - 'host.docker.internal:host-gateway'
    dns:
      - 1.1.1.1

    # storage
    volumes:
      - db-data:/var/lib/data
      - ./src:/app/src:ro
    tmpfs:
      - /tmp

    # ordering and health
    depends_on:
      db:
        condition: service_healthy
    healthcheck:
      test: ['CMD', 'curl', '-f', 'http://localhost/health']
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 20s

    # host access and limits
    cap_add:
      - SYS_PTRACE
    devices:
      - /dev/dri:/dev/dri
    ulimits:
      nofile: 65535
    deploy:
      resources:
        limits:
          cpus: '1.5'
          memory: 512M

    # bookkeeping
    labels:
      com.example.team: platform
    logging:
      driver: json-file
      options:
        max-size: '10m'
        max-file: '3'

🏗 Build Options

services:
  app:
    build:
      context: .
      dockerfile: docker/Dockerfile
      target: production
      args:
        NODE_VERSION: '22'
      cache_from:
        - myapp:cache
      secrets:
        - npmrc
      platforms:
        - linux/amd64
        - linux/arm64
    image: myapp:1.2 # the name given to the built image

📦 Volumes

services:
  db:
    volumes:
      # named volume — managed by Docker, survives 'down'
      - db-data:/var/lib/postgresql/data
      # bind mount — a host path, read-only
      - ./config:/etc/app:ro
      # anonymous volume — keeps node_modules out of the bind mount above
      - /app/node_modules
      # long syntax
      - type: bind
        source: ./src
        target: /app/src
        read_only: true

volumes:
  db-data:
  shared:
    external: true # created outside Compose
  nfs-data:
    driver_opts:
      type: nfs
      o: addr=10.0.0.10,rw
      device: ':/exports/data'

🌐 Networks

services:
  web:
    networks:
      - frontend
  api:
    networks:
      frontend:
        aliases:
          - api.internal
      backend:

networks:
  frontend:
    driver: bridge
  backend:
    internal: true # no outbound access
  existing:
    external: true
    name: some-other-network

💡 Compose creates a default network per project and every service joins it, so containers already reach each other by service name. Declare networks when you want to separate things.

🔀 Ports

services:
  web:
    ports:
      - '8080:80' # host:container
      - '127.0.0.1:8080:80' # bind to one interface only
      - '8080-8090:80-90' # a range
      - '80' # random host port
      - target: 80 # long syntax
        published: '8080'
        protocol: tcp
        mode: host

🧬 Environment Variables

services:
  app:
    environment:
      # map form (preferred)
      LOG_LEVEL: debug
      # take the value from the shell or .env
      API_KEY: ${API_KEY}
      # with a default, and a hard requirement
      PORT: ${PORT:-3000}
      DB_URL: ${DB_URL:?DB_URL must be set}
    env_file:
      - path: .env.production
        required: false

# variables come from the shell, .env, and --env-file, in that order of precedence
API_KEY=abc docker compose up -d
docker compose --env-file .env.staging up -d

🩺 Healthchecks & Startup Order

depends_on on its own only waits for the container to start. Wait for it to be healthy instead:

services:
  db:
    image: postgres:16
    healthcheck:
      test: ['CMD-SHELL', 'pg_isready -U postgres']
      interval: 10s
      timeout: 5s
      retries: 5
      start_period: 30s

  api:
    image: myapi
    depends_on:
      db:
        condition: service_healthy # or service_started, service_completed_successfully

🔄 Restart Policies

🧩 Profiles

Profiles keep optional services out of the way until you ask for them.

services:
  web:
    image: nginx # no profile: always started

  debug:
    image: busybox
    command: top
    profiles: [debug]

  seed:
    image: myapp
    command: npm run seed
    profiles: [tools]

# start the default services plus one profile
docker compose --profile debug up -d

# several at once
docker compose --profile debug --profile tools up -d

# COMPOSE_PROFILES works too
COMPOSE_PROFILES=debug,tools docker compose up -d

👀 Watch Mode

docker compose watch syncs or rebuilds automatically as you edit — a dev loop without bind-mount surprises.

services:
  web:
    build: .
    develop:
      watch:
        - action: sync
          path: ./src
          target: /app/src
        - action: rebuild
          path: package.json
        - action: sync+restart
          path: ./config
          target: /etc/app

docker compose watch

♻️ Reuse: include, extends & anchors

# pull in another compose file as if it were written here
include:
  - path: ./monitoring/compose.yaml

services:
  # inherit another service's definition
  worker:
    extends:
      file: common.yaml
      service: base-app
    command: ["node", "worker.js"]

# YAML anchors for repeated blocks
x-logging: &default-logging
  driver: json-file
  options:
    max-size: "10m"

services:
  api:
    logging: *default-logging
  web:
    logging: *default-logging

🔐 Secrets & Configs

File-based secrets work in plain Compose: each one is mounted read-only at /run/secrets/<name>. External secrets and configs require Swarm.

services:
  db:
    image: postgres:16
    environment:
      POSTGRES_PASSWORD_FILE: /run/secrets/db_password
    secrets:
      - db_password
    configs:
      - source: pg_conf
        target: /etc/postgresql/postgresql.conf

secrets:
  db_password:
    file: ./db_password.txt
  api_token:
    external: true # Swarm only

configs:
  pg_conf:
    file: ./postgresql.conf

🖥 GPU & Device Access

services:
  # NVIDIA — needs the NVIDIA Container Toolkit on the host
  cuda-app:
    image: nvidia/cuda:12.4.1-base-ubuntu22.04
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all # or: device_ids: ["0"]
              capabilities: [gpu]

  # Intel iGPU — VAAPI/OpenCL through the render device
  igpu-app:
    image: intel/openvino
    devices:
      - /dev/dri:/dev/dri
    group_add:
      - video

🚀 Everyday Commands

# start everything in the background
docker compose up -d

# rebuild the images first
docker compose up -d --build

# start one service and what it depends on
docker compose up -d <service>

# recreate containers even if nothing changed
docker compose up -d --force-recreate

# stop and remove containers and networks
docker compose down

# also remove the named volumes and local images
docker compose down -v --rmi local

# what is running, including health
docker compose ps

# follow the logs
docker compose logs -f
docker compose logs -f --tail 100 <service>

# a shell inside a running service
docker compose exec <service> sh

# a one-off container for a task
docker compose run --rm <service> npm test

# rebuild, pull, restart
docker compose build --no-cache
docker compose pull
docker compose restart <service>

# scale a stateless service
docker compose up -d --scale worker=3

# validate and print the effective configuration
docker compose config
docker compose config --services

# processes and resource usage
docker compose top
docker compose stats

⚔️ Compose vs Swarm Stacks

Both read a Compose file, but docker stack deploy honours a different subset of it.

🧭 Keep compose.yaml for local development and a separate stack.yaml for Swarm, or put the Swarm-only bits in an override file.

📚 Resources

• Compose file reference
• docker compose CLI reference
• Compose watch
• Environment variables and interpolation
• NVIDIA Container Toolkit