Backups

Automatic, out of the box: the backup container runs pg_dump nightly (03:00 UTC, configurable via BACKUP_HOUR) into a named Docker volume, in PostgreSQL’s compressed custom format, keeping 7 days (BACKUP_KEEP_DAYS).

docker compose exec backup ls -lh /backups          # see your backups
docker compose exec backup sh /backup.sh once       # take one right now

Keep off-box copies

A backup on the same disk as the database doesn’t survive the disk. Set an S3-compatible bucket (AWS, DigitalOcean Spaces, Cloudflare R2) in .env and the scheduler mirrors every dump daily:

BACKUP_S3_BUCKET=your-bucket
BACKUP_S3_PREFIX=db-backups
AWS_ACCESS_KEY_ID=...               # already in .env for media storage
AWS_SECRET_ACCESS_KEY=...

then docker compose up -d scheduler. Media files live in the media volume — back that up too, or switch media to S3 (DJANGO_DEFAULT_FILE_STORAGE=s3).

Restoring a backup

Practice this once on day one — a backup you’ve never restored is a hope, not a backup. (The shipped test suite does exactly this drill in CI.)

Drill (into a scratch database, zero risk):

docker compose exec backup createdb restore_drill
docker compose exec backup sh -c 'pg_restore -d restore_drill --no-owner /backups/<FILE>.dump'
docker compose exec backup psql -d restore_drill -c '\dt' | head   # tables are there
docker compose exec backup dropdb restore_drill

Real restore (replaces live data — stop the app first):

docker compose stop web scheduler
docker compose exec backup sh -c 'pg_restore -d $POSTGRES_DB --clean --if-exists --no-owner /backups/<FILE>.dump'
docker compose start web scheduler
curl -s https://api.your-domain.com/health/

Disaster recovery, summarized

1. New VPS (either runbook, steps 1–5).
2. Copy your .env (you kept a safe copy — right? If not: a new secret key invalidates all sessions/tokens, but your data is fine).
3. docker compose up -d, wait healthy.
4. Restore the latest dump from your S3 bucket (real-restore steps above).
5. Re-point DNS at the new IP.

Tip

Do the restore drill on day one, then once a quarter. The whole point of a nightly backup is a boring recovery — and the only way to know it’s boring is to have done it before you needed it.

Related: Upgrading (take a backup before every upgrade) · Hardening.