[BUG] Docker compose up --watch stopped working

[tamis-laan]

Description

I upgraded to the latest version of docker compose and now I get the following:

              ⦿ Syncing service "app" after 1 changes were detected
app-1         | INFO:uvicorn.error: Shutting down
app-1         | INFO:uvicorn.error: Waiting for application shutdown.
app-1         | INFO:uvicorn.error: Application shutdown complete.
app-1         | INFO:uvicorn.error: Finished server process [1]
              ⦿ service(s) ["app"] restarted
app-1 exited with code 0

I have sync+restart in my watch configuration for app. On file change the app is properly restarted but the logs and not printed any more. I know this because docker compose logs -f app will show logs properly.

Now this problem has come up in the past and was fixed, but I think someone reverted things or reimplemented the same mistake.

I'm on macos 14.7.6 (23H626) using the following versions through homebrew:

colima 0.8.4
docker 28.3.3
docker-buildx 0.26.1
docker-compose 2.39.2

[mateuspiresl]

I've been experiencing similar behavior since last week.

Since then, restarting a container stops displaying the service logs in the existing compose up process output.

I've heard colleagues experiencing the same and also seeing duplicated log lines. I couldn't reproduce that last behavior, though.

[coredumperror]

Same thing happens to me on Docker Desktop 4.45.0 with Docker Compose v2.39.2-desktop.1.

It does not happen with Docker Desktop 4.44.3 and Docker Compose v2.39.1-desktop.1, so it seems to have been introduced in the latest patch release of Compose (which is the version that fixed the log duplication... so I bet that fix is probably the cause).

As an additional note, pressing Ctrl-C on a docker compose up --watch with v2.39.2-desktop.1 immediately kills the app, rather than starting the container shutdown sequence, which is what it did before. Possibly related?

[ulyssessouza]

@tamis-laan Could you please provide a minimalistic version of your compose.yaml so that I can reproduce the issue?

[ndeloof]

I was not able to reproduce with a minimal example:

services:
  app:
    build:
      dockerfile_inline: |
        FROM alpine:latest
        CMD ping localhost
    develop:
      watch:
        - path: .
          target: /tmp
          action: sync+restart

@coredumperror can't reproduce either:

$ docker compose up --watch
[+] Running 1/1
 ✔ Container truc-app-1  Running                                                                                                                                                                                          0.0s 
        ⦿ Watch enabled
Attaching to app-1
app-1   | 64 bytes from ::1: seq=26 ttl=64 time=0.281 ms
app-1   | 64 bytes from ::1: seq=27 ttl=64 time=0.206 ms
Gracefully Stopping... press Ctrl+C again to force
 Container truc-app-1  Stopping
app-1   | 64 bytes from ::1: seq=28 ttl=64 time=0.099 ms
...
Container truc-app-1  Stopped

[coredumperror]

Could it be related to the fact that I have two containers running the same image, with the same files attached to sync+restart rules? They use different env vars to replicate the two different servers that we run the same software on.

I also have a mysql:8.4, elasticsearch:7.10.1, redis/redis-stack:latest, and axllent/mailpit container in my docker-compose.yml file.

Here's a somewhat stripped down and anonymized version version of our docker-compose.yml for the two custom services:

  multitenant:
    image: sites:latest
    container_name: multitenant
    hostname: multitenant
    networks:
      mtnet:
        ipv4_address: 172.20.0.10
    cap_add:
      # We need SYS_PTRACE to run py-spy to profile our tests
      - SYS_PTRACE
    ports:
      # Map port 443 on the host to port 8443 in the container, allowing access via normal https:// urls.
      - "443:8443"
      # The port that the debug server for sites listens on, for VS Code to connect to it in "connect" mode.
      - "6789:6789"
    env_file:
      - /etc/context.d/multitenant.env
      - /etc/context.d/aws_sso_credentials.env
    depends_on:
      - mysql
      - elastic
      - redis
    extra_hosts:
      # Maps DNS hostnames to the actual IPs of the appropriate containers. Required for cross-container syncs.
      - "main.localhost:172.20.0.11"
      - "pma.localhost:172.20.0.10"
    build:
      context: .
      tags:
        - sites:6.13.5
        - sites:latest
    develop:
      # Create a `watch` configuration to update the app's code in the image as it changes on the host.
      watch:
        # Sync the working directory with the /app directory in the container.
        - action: sync
          path: .
          target: /app
          ignore:
            # Ignore all migrations directories, because those are controlled by the volume mounts below.
            - "**/migrations"
        # Optional: Automatically detect changes to config files and restart the service to pick them up.
        # This is especially nice because you don't have to rebuild the image to activate the changes.
        - action: sync+restart
          path: ./etc/nginx/nginx.conf
          target: /app/etc/nginx/nginx.conf
          path: ./etc/supervisord.conf
          target: /app/etc/supervisord.conf
        # Optional: Automatically rebuild the image (and restart the service) upon any change to these files.
        - action: rebuild
          path: ./uv.lock
        - action: rebuild
          path: ./.dockerignore
        - action: rebuild
          path: ./Dockerfile
    volumes:
      # We use volume mounts for migrations directories so that `manage.py makemigrations` can write to the repo.
      - ./multitenant/api/migrations:/app/multitenant/api/migrations
      - ./multitenant/core/migrations:/app/multitenant/core/migrations
      - ./multitenant/redirects/migrations:/app/multitenant/redirects/migrations

  multitenant_www:
    image: sites:latest
    container_name: multitenant_www
    hostname: multitenant_www
    networks:
      mtnet:
        ipv4_address: 172.20.0.11
    cap_add:
      # We need SYS_PTRACE to run py-spy to profile our tests
      - SYS_PTRACE
    ports:
      # Map port 8443 on the host to port 8443 in the container, allowing access at https://main.localhost:8443
      - "8443:8443"
      # The port that the debug server for www listens on, for VS Code to connect to it in "connect" mode.
      - "6790:6790"
    env_file:
      - /etc/context.d/multitenant_www.env
      - /etc/context.d/aws_sso_credentials.env
    depends_on:
      - mysql
      - elastic
      - redis
    build:
      context: .
      tags:
        - sites:6.13.5
        - sites:latest
    develop:
      # Create a `watch` configuration to update the app's code in the image as it changes on the host.
      watch:
        # Sync the working directory with the /app directory in the container.
        - action: sync
          path: .
          target: /app
          ignore:
            # Ignore all migrations directories, because those are controlled by the volume mounts below.
            - "**/migrations"
        # Optional: Automatically detect changes to config files and restart the service to pick them up.
        # This is especially nice because you don't have to rebuild the image to activate the changes.
        - action: sync+restart
          path: ./etc/nginx/nginx.conf
          target: /app/etc/nginx/nginx.conf
        - action: sync+restart
          path: ./etc/supervisord.conf
          target: /app/etc/supervisord.conf
        # Optional: Automatically rebuild the image (and restart the service) upon any change to these files.
        - action: rebuild
          path: ./uv.lock
        - action: rebuild
          path: ./.dockerignore
        - action: rebuild
          path: ./Dockerfile
    volumes:
      # We use volume mounts for migrations directories so that `manage.py makemigrations` can write to the repo.
      - ./multitenant/api/migrations:/app/multitenant/api/migrations
      - ./multitenant/core/migrations:/app/multitenant/core/migrations
      - ./multitenant/redirects/migrations:/app/multitenant/redirects/migrations