docker_compose restarts even initially created containers

[sistason]

Summary

I am unsure if it's a bug or just an (for me) unexpected behaviour:

When using docker_compose with restarted=yes, the freshly container gets restarted right after creation, often breaking entrypoints and such.

Issue Type

Bug Report

Component Name

docker_compose

Ansible Version

2.11.1 with python 3.10

Community.general Version

community.general 7.0.1

Configuration

not applicable

OS / Environment

debian bullseye

Steps to Reproduce

- copy:
    dest: /srv/foo/docker-compose.yml
    content: |
      version: "3"
      services:
        db:
          image: mysql:5.7
          environment:
            MYSQL_DATABASE: foo
            MYSQL_ROOT_PASSWORD: "bar"
  register: _composefile
- docker_compose:
    project_src: "/srv/foo"
    project_name: "foo"
    state: "present"
    restarted: "{{ _composefile.changed }}"

Expected Results

I expected restarted=yes to not restart containers which were just created, as the whole point of restarting is that the container runs anew with all previously changed things (mostly changed mounted configfiles, which compose does not know about).

Actual Results

In the example, the mysql entrypoint creating the table gets interrupted, leading to a broken mysql. But I'd like to keep the restart functionality, depending on changed config/compose files.

handlers have the same problem, but the other way around - the handler needs restarted: yes, but shouldn't run initially. Do I really have to find out if the container did not exist before, and guard the restarted against that?

I'd suggest restarted does not restart services which it just now started.

Code of Conduct

• I agree to follow the Ansible Code of Conduct

[felixfontein]

Good question. The documentation of the state option isn't fully clear whether docker-compose up is ran before docker-compose stop or docker-compose restart or not - the implementation does exactly that, though. I guess part of it is because the implementation wants to make sure that the service is actually in a correct state (i.e. all containers are up and running with the correct configuration etc.) before restating. Changing this in a sensible way is not so easy, I guess.

[jacobw]

I have a similar issue.

I use a handler with restarted: true which I expect to start (up) the containers if they are not running already (usually on first deploy), and restart them if they are already running. This is breaking a project that needs to do some things on first run only.