up --wait does not fail when a service becomes unhealthy

[wakairo]

Describe the bug

podman-compose up --wait does not fail when a service becomes unhealthy.

When a service uses depends_on with condition: service_healthy, up --wait continues waiting after the dependency has become unhealthy. This wait can also continue beyond --wait-timeout.

Without the dependency, --wait-timeout ends the wait but returns status 0 even though the service remains unhealthy.

Docker Compose reports an error when a required service becomes unhealthy. podman-compose should provide compatible behavior so that CI and scripts neither hang nor report false success.

To Reproduce

Create a directory for the reproducer:

mkdir -p /tmp/podman-compose-unhealthy-repro
cd /tmp/podman-compose-unhealthy-repro

Create compose.yaml:

services:
  dependency:
    image: busybox:latest
    command: ["sh", "-c", "while :; do sleep 3600; done"]
    healthcheck:
      test: ["CMD-SHELL", "exit 1"]
      interval: 1s
      timeout: 1s
      retries: 1

  app:
    image: busybox:latest
    command: ["sh", "-c", "while :; do sleep 3600; done"]
    depends_on:
      dependency:
        condition: service_healthy

The commands below explicitly use unhealthy-repro as the Compose project name:

timeout 10s podman-compose -p unhealthy-repro \
  up -d --wait --wait-timeout 3

printf 'exit status: %s\n' "$?"

Inspect the containers:

podman ps -a \
  --filter label=com.docker.compose.project=unhealthy-repro \
  --format '{{.Names}} {{.Status}}'

Clean up:

podman-compose -p unhealthy-repro down

Expected behavior

As soon as dependency becomes unhealthy:

• podman-compose up -d --wait should stop waiting.
• The command should return a non-zero status.
• app should not be treated as ready.
• --wait-timeout should bound all readiness waits, including service_healthy dependency waits.

This matches the behavior of Docker Compose.

Actual behavior

The dependency becomes unhealthy, but podman-compose continues waiting.

The external timeout command eventually terminates podman-compose:

exit status: 124

--wait-timeout 3 does not bound this dependency wait.

As a related case, if the app service is removed so that only the unhealthy service remains, this command returns status 0 after the timeout even though the service is still unhealthy:

podman-compose -p unhealthy-repro up -d --wait --wait-timeout 3

Output

$ podman-compose version
podman-compose version 1.6.0

$ podman --version
podman version 4.9.3

Environment:

• OS: Ubuntu 24.04
• Podman: 4.9.3, rootless
• podman-compose: 1.6.0

Additional context

In podman-compose 1.6.0, the readiness code waits for the healthy condition but does not appear to treat unhealthy as a failure:

https://github.com/containers/podman-compose/blob/v1.6.0/podman_compose.py#L3493-L3535

The service_healthy dependency wait also does not appear to be covered by --wait-timeout:

https://github.com/containers/podman-compose/blob/v1.6.0/podman_compose.py#L3685-L3730

Docker Compose explicitly reports an error when a service becomes unhealthy:

https://github.com/docker/compose/blob/main/pkg/compose/convergence.go

A possible fix would be to:

1. Wait for both healthy and terminal failure states such as unhealthy.
2. Return a non-zero status immediately when a required service becomes unhealthy.
3. Apply --wait-timeout to dependency-condition waits as well as the final readiness wait.