optimize plex sync: docker exec + single async loop + version-controlled crontab

[steezeburger]

Follow-up to #34. Three changes that together address the OOM-kills-bot problem caused by the every-10-minute plex sync.

Summary

1. bin/dcp-django-admin.sh prefers docker compose exec against the running web container instead of docker compose run --rm. Each invocation no longer spins up a brand-new ephemeral container (~100–150 MB peak on this image), which is what was pushing the host into OOM territory and taking the bot down. Falls back to compose run --rm when web isn't running (e.g. during the very first migrate). Also reads POSTGRES_HOST from .env directly instead of spawning a container just to inspect env, and skips the confirmation prompt when stdin isn't a TTY (so cron / GitHub Actions / SSH don't hang).

2. SyncWithPlexCommand async refactor. Phase 1 walks Plex movies and persists them; Phase 2 runs ONE asyncio.run over all enrichments inside a single shared aiohttp.ClientSession via asyncio.gather. The previous code spun up a fresh event loop and ClientSession per movie. EnrichMovieActorsCommand.execute now takes an optional session= parameter so the existing manual backfill (manage.py enrich_actors) keeps working unchanged.

3. crontab.txt + just install-crontab. Commits the existing schedule to the repo and adds a recipe to install it. ⚠️ See open question below — this is host-side cron, which we may want to revisit.

Open question: host cron vs cron-in-docker

What's in this PR is host cron — crontab.txt is installed onto the VPS via just install-crontab and the system cron daemon still runs the schedule. It only version-controls the entries.

After committing this, we agreed the better shape is cron in docker — either a dedicated scheduler service running cron -f (mirrors brainspread's brainspread-scheduler-1) or cron inside the existing web container. That's not in this PR and would be a separate change that drops crontab.txt/install-crontab and adds the chosen approach.

Two ways to land this:

• Merge as-is — get the docker-exec + async-refactor wins immediately; remove the host-cron pieces in a follow-up alongside introducing the scheduler service.
• Strip the crontab pieces from this PR — keep the description tight, ship docker-exec + async refactor only, and tackle scheduling holistically next.

Either is fine — flagging so you can decide.

What changed (4 files)

• app/plex/commands.py — async refactor of SyncWithPlexCommand + optional session= on EnrichMovieActorsCommand.execute
• bin/dcp-django-admin.sh — compose exec preferred over compose run, non-TTY-safe, reads env from .env
• crontab.txt — new, byte-identical to what's currently installed on the VPS
• justfile — adds install-crontab recipe

Deploy steps

The auto-deploy added in #34 should run this entire PR on merge. After it lands:

1. Confirm services are healthy:

   docker ps --format 'table {{.Names}}\t{{.Status}}'

2. Diagnose enrichment backfill scope. The OOM kills mid-sync caused recent movies to never get enriched — the next sync's latest_movie early-exit then skips them permanently:

   docker compose exec -T web python manage.py shell -c "
   from plex.models import PlexMovie
   total = PlexMovie.objects.count()
   enriched = PlexMovie.objects.filter(actors_enriched_at__isnull=False).count()
   print(f'enriched: {enriched}/{total} ({100*enriched/max(total,1):.1f}%)')
   "

3. Backfill if the percentage is low:

   docker compose exec -T web python manage.py enrich_actors

4. (Optional, only if keeping the host-cron pieces) Install the version-controlled crontab — functionally a no-op since entries are byte-identical, but moves source-of-truth to the repo:

   cd /root/oscarr-stuff/oscarr/packages/django-app
   just install-crontab

Test plan

• Lint and Test passes
• Deploy to Production fires automatically and SSHes into the VPS successfully
• After merge: oscarr-web-1, oscarr-bot-1, oscarr-db-1 all Up
• just sync-with-plex does not spawn a new container (verify via docker events --filter event=create during a sync)
• Cron-driven sync at the next */10 * * * * boundary completes without prompt-hang and without OOM
• manage.py enrich_actors still works (signature compat for the manual backfill)

https://claude.ai/code/session_01RJrZb4X5Ywuzq2SjeFXgN1