fix(site_worker): remove watchmedo auto-restart causing BrokenPipeError loop

[AybH26]

fix(site_worker): remove watchmedo wrapper that caused restart loop (#2417)

Closes #2417

Issue

Uploading a competition bundle leaves the "unpacking" step spinning forever. The unpack_competition Celery task is enqueued on the site-worker queue but never executed, because the site_worker container — although reported as Up by Docker — has no consumer attached to RabbitMQ.

Root cause

The previous site_worker command wrapped Celery in watchmedo so that task code edits would auto-reload:

command: ["watchmedo auto-restart -p '*.py' --recursive -- celery -A celery_config worker -B -Q site-worker -l info -n site-worker@%n --concurrency=2"]

watchmedo watches the entire mounted /app tree, which includes:

• __pycache__/*.pyc files written by Python on every import
• src/celerybeat-schedule* SQLite files written continuously by Celery beat itself

These files keep changing faster than Celery can finish initializing its prefork pool. Each aborted startup leaks pool worker processes and produces a BrokenPipeError in billiard. As a result no consumer ever attaches to the site-worker queue and tasks pile up indefinitely.

Fix

Remove the watchmedo wrapper from the site_worker command in docker-compose.yml:

   site_worker:
-    # This auto-reloads
-    command: ["watchmedo auto-restart -p '*.py' --recursive -- celery -A celery_config worker -B -Q site-worker -l info -n site-worker@%n --concurrency=2"]
+    command: ["celery -A celery_config worker -B -Q site-worker -l info -n site-worker@%n --concurrency=2"]
     working_dir: /app/src