fix: update depends_on to use Compose v2 condition clauses

Andrew Dunnrequested to merge
fix-compose-dependency-ordering into master

What does this MR do?

Fix the documented "down and re-up" webservice startup flakiness by updating depends_on to use Compose v2 condition clauses, matching the CNG upstream pattern.

Problem

The README documents a known workaround:

You likely will have to down and re-up to get the webservice container to start properly.

This happens because docker-compose.yml uses bare service names in depends_on without health conditions. Webservice starts before postgres is healthy and before migrations complete. The internal wait-for-deps script has a 60-second timeout, but if migrations take longer, webservice fails.

The CNG upstream docker-compose.yml has already been updated to use Compose v2 long-form depends_on with condition: service_healthy and condition: service_completed_successfully. This MR brings dsop-scripts in line with that pattern.

Changes

  1. Remove version: "3": Compose v2 does not require a version field, and the condition: syntax is a v2 feature.

  2. Add healthchecks to postgres and redis, matching the CNG upstream pattern (pg_isready for postgres, redis-cli ping for redis).

  3. Update all depends_on blocks to use conditions:

Service Waits for
migrations postgres: healthy, redis: healthy
webservice postgres: healthy, redis: healthy, migrations: completed
workhorse webservice: healthy, redis: healthy
sidekiq postgres: healthy, redis: healthy, migrations: completed
pages webservice: healthy
registry_migrations postgres: healthy
registry registry_migrations: completed
kas redis: healthy
  1. Add KAS websocket secret volume mount, prerequisite from the companion CNG MR.

Notes

  • Compose v2 required. The condition: syntax requires Docker Compose v2 (the docker compose plugin, v2.17+). Legacy docker-compose v1 does not support conditions.
  • Webservice healthcheck is defined in the CNG image (HEALTHCHECK instruction), not in the compose file. condition: service_healthy resolves via the image-level healthcheck.
  • Depends on CNG MR for the KAS websocket secret file. Merge CNG first.
  • README update: The "down and re-up" workaround can be removed once this MR is merged. Happy to include that here or as a follow-up.

Related issues

#107

Test plan

Tested 3 times from clean state (docker compose down -v) on CentOS Stream 10 (GCP, Docker CE 29.3.1, Compose v5.1.1) with CNG images v18.10.0:

Run Result Duration Webservice first-try?
1 PASSED 218s Yes
2 PASSED 216s Yes
3 All containers healthy ~210s Yes

Webservice started successfully on the first docker compose up every time. No "down and re-up" needed.

AI-Generated Content Disclosure: This MR contains code generated with assistance from GitLab Duo and OpenCode. The output has been reviewed for correctness, tested, and validated against project requirements per GitLab's AI contribution guidelines.

Merge request reports