Production Deployment

Production Deployment (db.zizka.ai)

How the managed cloud instance is deployed on EC2 without losing user data.

Golden rule

Never run docker compose down -v on production.

The -v flag deletes the Postgres volume — all signups, tenants, API keys, and events are gone permanently unless you restore from backup.

Task	Command
Production deploy	bash infra/deploy-production.sh
Backup only	bash infra/backup-postgres.sh
Qdrant backup	bash infra/backup-qdrant.sh
Local fresh install	bash scripts/reset-local-db.sh (laptop only)

Safe production deploy (EC2)

cd ~/agentdb   # or your clone path
bash infra/deploy-production.sh

This script:

1. Backs up Postgres to infra/backups/zizkadb_<timestamp>.sql.gz
2. Runs git pull
3. Rebuilds API with docker compose up -d --build — no volume wipe
4. Restarts API and deploys dashboard
5. Prints user count before/after and fails if count drops to zero

Non-interactive:

ZIZKA_DEPLOY_CONFIRM=yes bash infra/deploy-production.sh

Daily backups (cron on EC2)

crontab -e

Add:

0 3 * * * cd /home/ubuntu/agentdb && bash infra/backup-postgres.sh >> /var/log/zizkadb-backup.log 2>&1

Backups are kept for 14 days (BACKUP_KEEP_DAYS to override).

Restore from backup

cd ~/agentdb
# Pick latest backup
ls -lt infra/backups/

gunzip -c infra/backups/zizkadb_YYYYMMDDTHHMMSSZ.sql.gz | \
  docker compose -f infra/docker-compose.yml exec -T postgres \
  psql -U zizkadb -d zizkadb

For a full restore into an empty database, stop API first, restore, then restart.

Pull latest (manual — prefer deploy-production.sh)

cd ~/agentdb
bash infra/backup-postgres.sh
git pull origin main
docker compose -f infra/docker-compose.yml up -d --build
docker compose -f infra/docker-compose.yml restart api
bash infra/deploy-dashboard.sh

API (Python / FastAPI)

Schema changes apply on API startup via core/db/connection.py (idempotent ALTER / CREATE IF NOT EXISTS). Do not rely on re-running schema.sql — that only runs on a brand-new empty Postgres volume.

Verify:

curl -s https://db.zizka.ai/health
# {"status":"ok","version":"0.1.0"}

Dashboard (Next.js)

cd ~/agentdb
bash infra/deploy-dashboard.sh

Dashboard runs on port 3001 behind nginx at db.zizka.ai.

Common deploy mistakes

Mistake	What happens	Fix
docker compose down -v on EC2	All users deleted	Restore backup; use deploy-production.sh
npm run build from repo root	Build fails	cd dashboard first
API not restarted after pull	New routes missing	docker compose restart api
Schema fix + volume wipe	Data loss	Use migrations + backup, not -v

Environment (production)

• ENV=production in infra/.env
• DATABASE_URL — Postgres
• REDIS_URL, QDRANT_URL
• JWT_SECRET, JWT_REFRESH_SECRET
• TRIAL_DAYS — trial length (default 30)
• API_KEY_LIMITS_ENFORCED — enable Pro/Team API key caps (default false)
• NEXT_PUBLIC_API_URL=https://db.zizka.ai for dashboard build
• Do not set DEV_API_KEY in production

Verify after deploy

1. curl https://db.zizka.ai/health → ok
2. Check user count in deploy script output (should not drop)
3. Login to dashboard
4. Create agent → Test agent → event appears

Qdrant backup and restore

Backup:

bash infra/backup-qdrant.sh

• Docker: creates a collection snapshot via Qdrant API → infra/backups/qdrant_docker_<timestamp>.snapshot
• Native: tar archive of .local/qdrant/storage/ → infra/backups/qdrant_native_<timestamp>.tar.gz

Restore (native):

bash scripts/stop-native-stack.sh
rm -rf .local/qdrant/storage
mkdir -p .local/qdrant/storage
tar -xzf infra/backups/qdrant_native_<timestamp>.tar.gz -C .local/qdrant
bash scripts/restart-native-stack.sh

Restore (Docker): use Qdrant snapshot upload API or restore the qdrant_data volume from backup per Qdrant docs.