Deploy

Frontend and docs deploy automatically on Vercel when main changes. This page covers the backend, which is deployed by hand.

1. Pre-checksCopy Anchor Link

ssh vod              # SSH config alias on the admin's machine
cd ~/vod-app/infra
alias dc='docker compose -f docker-compose.yml --env-file prod.env'

dc exec redis sh -c 'redis-cli -a "$REDIS_PASSWORD" --no-auth-warning GET video_processing:active_lock'

cd ~/vod-app
git fetch && git log --oneline HEAD..origin/main      # incoming commits
git diff --stat HEAD origin/main -- backend/alembic   # any new migrations?
git rev-parse --short HEAD                            # note this for rollback

2. ReleaseCopy Anchor Link

cd ~/vod-app && git pull origin main
cd infra
dc up -d --build api worker

Only api and worker are rebuilt; infrastructure containers are untouched unless their configuration changed. If infra/caddy/Caddyfile.prod changed, also run dc restart caddy.

MigrationsCopy Anchor Link

If the release contains new files in backend/alembic/versions/:

dc exec api alembic upgrade head
dc exec api alembic current        # should print the new head revision

The API's startup create_all() creates new tables but never alters existing ones, so skipping this breaks any release that changes a column.

3. VerifyCopy Anchor Link

dc ps                                                    # api and worker Up
curl -s https://api.vod.spokhrel.dev/health/             # {"status":"Ok!"}
dc logs --since 5m api | tail -20                        # no tracebacks
dc logs --since 5m worker | grep -E "ready|\[queue\]"    # "celery@… ready" and the recovery line

On startup the worker logs one line such as:

[queue] Worker startup recovery: purged 0 message(s), marked 0 interrupted video(s) failed: []

If it marked a video failed, re-queue it (Debugging).

Finally, open the site, play a video and, if the release touched the pipeline, upload a short test video and watch it reach completed.

4. Roll backCopy Anchor Link

cd ~/vod-app
git checkout <previous-commit>        # the hash noted in pre-checks
cd infra && dc up -d --build api worker

If the bad release ran a migration, downgrade it first while the new code is still checked out:

dc exec api alembic downgrade -1

Return to main with git checkout main once the fix is pushed.

What a deploy interruptsCopy Anchor Link