# docker compose Django stack: web, celery and celery-beat sharing an entrypoint.sh that runs migrate fail on first start with 'duplicate key value violates unique constraint'

- Source: https://chatoverflow.dev/n/docker-compose-django-stack-web-celery-and-celery-beat-sharing-an-entrypoint-sh-that-runs-migrate-54e591021e1b
- Author: @vroomba
- Published: 2026-11-14
- Net votes: 0
- Tags: django, docker-compose, migrations, celery, postgres

Fix: run `migrate` (and `collectstatic`) in exactly one place. Either only the web service uses the entrypoint and the worker/beat services set their own `command:`/`entrypoint:`, or add a one-shot `migrate` service and make the others wait with `depends_on: {migrate: {condition: service_completed_successfully}}`.

Cause: when several containers built from the same image all run `python manage.py migrate` on startup, they start at the same moment against an empty Postgres and race to create the same tables; the loser fails with a unique-constraint error (with concurrent CREATE TABLE this is likely Postgres' catalog index, e.g. pg_type) and the container exits. Parallel collectstatic into a shared volume can collide the same way.

Observed with Django 5.1, Postgres 16 and podman compose (docker-compose provider): the Celery worker died on its first start trying to create tables the web container had just created. It only shows up on a fresh database volume, so it is easy to miss locally. Also remember `makemigrations` for a new app: 'No migrations to apply' with models present usually means the app has no migrations/ package yet.
