compose depends_on attribute not respected

[mamasch19]

Hi, I am having issues with watchtower not respecting the dependency relationship set with the depends_on: attribute.
This is the compose of the stack I used for testing:

services:
  app:
    image: quay.io/pussthecatorg/breezewiki
    labels:
      - com.centurylinklabs.watchtower.enable=true
    restart: unless-stopped
  mon:
    image: busybox
    depends_on:
      - app
    labels:
      - com.centurylinklabs.watchtower.enable=true
    restart: unless-stopped
    stdin_open: true
    tty: true
  unmon:
    image: busybox
    depends_on:
      - app
    restart: unless-stopped
    stdin_open: true
    tty: true

Watchtower is configured to run every day, but whenever I check with docker compose ps, the "app" container has a lower uptime that the 2 dependents:

NAME                IMAGE                                     COMMAND             SERVICE   CREATED        STATUS        PORTS
testtower-app-1     quay.io/pussthecatorg/breezewiki:latest   "racket dist.rkt"   app       11 hours ago   Up 11 hours   10416/tcp
testtower-mon-1     busybox                                   "sh"                mon       30 hours ago   Up 14 hours   
testtower-unmon-1   busybox                                   "sh"                unmon     30 hours ago   Up 14 hours   

I expected the "mon" and "unmon" containers to both be restarted whenever "app" is updated, so the uptime should be nearly identical.
The watchtower log has no mention at all of the dependent containers, just the one being updated. I also made sure not to enable rolling restarts.
Is this some misconfiguration on my side or an issue with how watchtower works?

This is my watchtower configuration for reference:

compose.yaml

name: watchtower
services:
  app:
    image: ghcr.io/nicholas-fedor/watchtower:1.19.0
    env_file:
      - ./.env
    environment:
      - TZ=Europe/Berlin
    labels:
      - diun.enable=true
      - diun.watch_repo=true
      - diun.include_tags=^\d+\.\d+\.\d+$
    restart: unless-stopped
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
networks: {}

.env

# General Options
WATCHTOWER_UPDATE_ON_START=false

# Scheduling & Polling
WATCHTOWER_SCHEDULE=0 17 06 * * *
WATCHTOWER_POLL_INTERVAL=""

# Container Management
WATCHTOWER_INCLUDE_STOPPED=true
#WATCHTOWER_ROLLING_RESTART=true
WATCHTOWER_CLEANUP=true
WATCHTOWER_REMOVE_VOLUMES=true
#WATCHTOWER_NO_PULL=true
WATCHTOWER_LABEL_ENABLE=true
WATCHTOWER_LABEL_TAKE_PRECEDENCE=true
WATCHTOWER_USE_COMPOSE_DEPENDS_ON=true

# Registry & Authentication
WATCHTOWER_WARN_ON_HEAD_FAILURE=always
#(default is auto)

# Lifecycle & Health
#WATCHTOWER_COOLDOWN_DELAY=2d
WATCHTOWER_TIMEOUT=90s

# Notifications
WATCHTOWER_NOTIFICATION_URL=ntfy://REDACTED
#WATCHTOWER_NO_STARTUP_MESSAGE=true
WATCHTOWER_NOTIFICATIONS_LEVEL=debug
#WATCHTOWER_NOTIFICATION_SKIP_TITLE=true
#WATCHTOWER_NOTIFICATION_SPLIT_BY_CONTAINER=true
#WATCHTOWER_NOTIFICATION_TEMPLATE=

Some help would be greately appreciated.

[janitor-of-code]

This seems to be working as expected judging from the docker compose docs:
https://docs.docker.com/reference/compose-file/services/#depends_on

I think you want the long syntax, i.e.

Long syntax

The long form syntax enables the configuration of additional fields that can't be expressed in the short form.

restart: When set to true Compose restarts this service after it updates the dependency service. This applies to an explicit restart controlled by a Compose operation, and excludes automated restart by the container runtime after the container dies. Introduced in Docker Compose version 2.17.0

In your example that would be:

services:
  app:
    image: quay.io/pussthecatorg/breezewiki
    labels:
      - com.centurylinklabs.watchtower.enable=true
    restart: unless-stopped
  mon:
    image: busybox
    depends_on:
      app:
        restart: true
    labels:
      - com.centurylinklabs.watchtower.enable=true
    restart: unless-stopped
    stdin_open: true
    tty: true
  unmon:
    image: busybox
    depends_on:
      app:
        restart: true
    restart: unless-stopped
    stdin_open: true
    tty: true

[mamasch19]

I also came across that option in my research, but as far as I could tell from the docs, watchtower does not support that field for automatic restarts and should restart the containers in the correct order anyway:

Watchtower does not support using this to control implicit restarts, because this is an explicit, opt-in feature that defaults to false when omitted, such as when using the short-form syntax.

I still added it to my test stack, just to try it out (had to also include condition: service_started). I'll report back tomorrow, when the next update cycle is finished.

[mamasch19]

Didn't work, still only the "app" container was restarted.

[janitor-of-code]

There was another ticket that gained more traction and led to a fix in version v1.20.2

Does that solve what you're experiencing as well?