Depends On

depends_on: controls the order Compose starts services. Fibe passes it through unchanged. Use it to:

• wait for a database to be healthy before starting an app,
• wait for a setup/migration job to complete before starting the web tier,
• wait for a config-producing service to start before others read from its filesystem.

Three conditions

services:
  web:
    depends_on:
      db:
        condition: service_healthy
      setup:
        condition: service_completed_successfully
      cache:
        condition: service_started

Condition	Means
service_started	Compose started the container (no readiness guarantee)
service_healthy	The dependent service's Compose healthcheck: is passing
service_completed_successfully	The dependent service has exited with status 0 (one-shot jobs)

Long-form vs short-form

Compose allows the short form depends_on: [a, b] — equivalent to condition: service_started for each. For Fibe templates, use the long form: most apps need service_healthy for databases.

Healthcheck on the dependency

For condition: service_healthy to work, the dependency must define a Compose healthcheck::

services:
  db:
    image: postgres:17.5
    healthcheck:
      test: ["CMD-SHELL", "pg_isready -U postgres"]
      interval: 5s
      timeout: 5s
      retries: 5
      start_period: 30s
  web:
    image: my-app
    depends_on:
      db:
        condition: service_healthy

This is the standard pattern for apps that need a database to be reachable before setup, web, or worker roles start.

Setup/migration pattern

A common pattern: a one-shot service that runs migrations, then exits. Other services wait for it:

services:
  setup:
    image: my-app
    command: bin/rails db:prepare
    restart: 'no'
    depends_on:
      db:
        condition: service_healthy

  web:
    image: my-app
    command: bin/rails server -b 0.0.0.0
    depends_on:
      setup:
        condition: service_completed_successfully
      db:
        condition: service_healthy

Use this for setup services that prepare the app, then exit. Web and worker services can then wait for setup to complete successfully.

DRY with anchors

Repeating the same depends_on: across services is wasteful. Anchor it:

x-app-deps: &app-deps
  db:
    condition: service_healthy
  cache:
    condition: service_started
  setup:
    condition: service_completed_successfully

services:
  web:
    depends_on: *app-deps
  jobs:
    depends_on: *app-deps

See recipe-anchors-and-aliases.

Cyclic depends_on

Compose rejects cycles. If a depends on b and b depends on a, restructure: usually one of them is wrong (use service discovery + retry logic in the app instead of startup ordering).

Job-mode and depends_on

In job-mode templates, depends_on works as in Compose. Watched services (fibe.gg/job_watch: "true") should NOT depend on each other's service_completed_successfully if they should run in parallel.

depends_on vs in-app retry

App code still must retry connections — Compose start order isn't a hard guarantee. depends_on reduces transient errors at first start; the app must handle reconnection over its lifetime anyway.

Use the app framework's normal database retry support. For Node/PG, use a connection-pool retry strategy. For Python, use the database driver's retry settings or a retry library.

Pitfalls

• service_healthy without a healthcheck: — Compose fails to start: "depends_on service_healthy is not configured". Always pair.
• Healthcheck too strict during boot — app retries forever; start_period should be generous.
• Compose v2 vs v3 syntax — long-form depends_on works on both. Avoid the v3 deprecation of the short form.
• Depending on a service that has been removed from the template — Compose error. Audit depends_on keys.
• Depending on a static service from a dynamic service's build — depends_on is runtime ordering only; build happens before any runtime services exist.

Related skills

recipe-anchors-and-aliases, recipe-strip-incompatible-keys, playbook-rails-app, playbook-postgres-app.