Honor the restart policy

Report a container that exits 0 under a non-restarting policy as completed, so a one-shot dependency’s service_completed_successfully condition can resolve.

# compose.yml
services:
  sablier:
    image: sablierapp/sablier:1.15.0
    command:
      - start
      - --provider.name=docker
      - --provider.docker.honor-restart-policy=true
      - --strategy.blocking.default-timeout=2m
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock

  app:
    image: sablierapp/mimic:v0.3.3
    labels:
      - "sablier.enable=true"
      - "sablier.group=demo"
    depends_on:
      init:
        condition: service_completed_successfully

  init:
    image: sablierapp/mimic:v0.3.3
    restart: "no"

When app is requested, Sablier starts the one-shot init container, waits until it reports completed, then starts app. On later requests init is already completed, so Sablier does not restart it.

    flowchart LR
    req["Request for app"]
    init["init runs<br/>restart: no"]
    exit["init exits 0"]
    completed["Sablier reports completed<br/>honor-restart-policy=true"]
    app["app starts<br/>service_completed_successfully met"]
    req --> init --> exit --> completed --> app
  

By default Sablier reports any container that exits with code 0 as stopped. With --provider.docker.honor-restart-policy=true, Sablier instead honors the container’s Docker restart policy when it exits successfully:

completed is distinct from ready: a ready container is running and serving traffic, whereas a completed container ran once and exited. A completed dependency satisfies a service_completed_successfully condition but never a service_healthy one. Docker normalizes an unset restart policy to no, so a container with no declared policy behaves like restart: "no".

When to use it

Use this when a managed service depends on a one-shot init or migration container with service_completed_successfully. Without the flag, the exited container is reported as stopped, the condition never resolves, and Sablier keeps restarting it on every poll.

Flags

• --provider.docker.honor-restart-policy: report a container that exits 0 under a non-restarting policy as completed.

See the runnable example.