Hosting & Infrastructure¶
This document describes the hosting infrastructure for the Freeze Design webshop, including provider choices, cost breakdown, and scaling options.
Current Infrastructure¶
Two live environments
Staging (https://staging.freezedesign.nl, behind basic auth, docker-compose.staging.yml)
and production (https://www.freezedesign.nl, docker-compose.prod.yml on a Hetzner CX43)
are both live. Production runs tagged releases from main, deployed with
scripts/deploy-production.sh (v1.60 on 2026-11-06); the shop went live on
2026-09-30. freezedesign.nl and prod.freezedesign.nl redirect to www. Each
environment is one VPS; /opt/webshop holds the compose file, env file, scripts and nginx
config that the deploy workflows copy there.
The table describes both environments.
| Component | Provider | Details |
|---|---|---|
| VPS | Hetzner Cloud (Nuremberg) | Staging: 8 GB RAM / 4 vCPU (PostgreSQL tuned with shared_buffers=512MB). Production: CX43, 16 GB RAM / 8 vCPU (docker-compose.prod.yml sums container mem_limits to roughly 3.75 GB) |
| Media storage | DigitalOcean Spaces | ams3, buckets freeze-production-media / freeze-staging-media, served straight from the bucket (MEDIA_URL points at <bucket>.ams3.digitaloceanspaces.com); versioning on since 2026-09-28, and scripts/backup-media.sh copies the bucket with rclone to a second bucket |
| Database backups | DigitalOcean Spaces | Hourly AES-256-GCM encrypted pg_dump uploads (BACKUP_S3_*, bucket freeze-production-backups on production); tiered retention 24 hourly / 7 daily / 4 weekly, see Disaster Recovery |
| SSL | Let's Encrypt | Free; renewed weekly by ssl-renewal.yml for both environments plus a root crontab on each VPS (see Disaster Recovery) |
| DNS | Openprovider (registrar) | freezedesign.nl, www, prod and staging point straight at the VPSes; no Cloudflare proxy |
| CDN / DDoS | none | nginx limit_req zones only. A Cloudflare front was designed (Cloudflare CDN) but not put in use; the production-readiness review (item II-2) advised closing it, because a nameserver move touches the DNSSEC configured at go-live |
| Monitoring | Sentry, PostHog, Discord | Sentry (backend + frontend projects per environment), PostHog analytics via /ingest (one project for both environments, environment property on every event), Discord webhooks for deploys, backups, payments, errors and the Informer bookkeeping link; external uptime monitoring is still an open todo (externe-uptime-monitor) |
| Datacenter | Germany | EU location for GDPR compliance |
VPS sizing
Checked on the servers on 2026-09-28: staging has 8 GB RAM and 4 vCPU, production
16 GB RAM and 8 vCPU. The compose files use far less than that: staging tunes
PostgreSQL to shared_buffers=512MB, and docker-compose.prod.yml sums container
mem_limits to roughly 3.75 GB (db 1 GB, backend 1 GB, celery 768 MB, frontend 512 MB,
redis 256 MB, celery-beat 256 MB; nginx has no limit). There is room to grow without a
bigger machine. scripts/provision-server.sh adds a 4 GB swap file as a safety net.
The application runs as 7 Docker containers orchestrated by Docker Compose (service names
from docker-compose.prod.yml; staging uses the same names):
- nginx (
nginx:1.31.6-alpine) -- reverse proxy, SSL termination, rate limiting, static file serving; config bind-mounted fromnginx/nginx.conf(nginx.staging.confon staging)./api/agent/only admits the Freeze AI VM's tailnet IP (on staging without basic auth); see agent API - backend (
ghcr.io/voorman/webshop_freeze_design/backend,python:3.14-slim) -- Django/Gunicorn on port 8000, REST API under/api/, Django admin under/django-admin/ - frontend (
ghcr.io/voorman/webshop_freeze_design/frontend,node:26-alpine) -- Next.js server-side rendering on port 3000, shop and/admin - db (
postgres:15-alpine) -- PostgreSQL 15 with SSL and per-environment tuning - redis (
redis:7-alpine) -- cache and Celery broker - celery -- asynchronous task worker (concurrency 2 on production, 1 on staging;
--max-memory-per-child=262144in both,mem_limit 768mon production since #990) - celery-beat -- periodic task scheduler (backups, payment checks, mails);
HeartbeatSchedulerwith a healthcheck, restarted by cron when unhealthy (scripts/restart-unhealthy.sh, see Monitoring)
flowchart LR
U[Browser] -->|HTTPS| DNS["DNS at Openprovider<br/>www / apex / prod / staging"]
DNS --> NG
subgraph VPS["Hetzner VPS (/opt/webshop, Docker Compose)"]
NG["nginx :80/:443<br/>TLS, redirects, rate limits"]
NG -->|/| FE["frontend<br/>Next.js :3000"]
NG -->|"/api/, /django-admin/, /_allauth/"| BE["backend<br/>Django/Gunicorn :8000"]
NG -->|"/static/"| ST[(static volume)]
FE -->|server-side fetch| BE
BE --> DB[(db<br/>PostgreSQL 15)]
BE --> RD[(redis)]
CW["celery worker"] --> DB
CW --> RD
CB["celery-beat"] --> RD
end
BE -->|media, S3 API| SP["DigitalOcean Spaces (ams3)<br/>media bucket"]
U -->|product images| SP
CW -->|hourly encrypted backups| BK["Spaces backup bucket"]
BE --> MO[Mollie]
BE --> SE[Sentry]
FE --> SE
FE -->|/ingest| PH[PostHog]
CW --> DC[Discord webhooks]
CB -. schedules .-> CW
Both environments have this shape; the k6 gate, E2E runs and the deploy workflows reach the
VPS over SSH from GitHub Actions (.github/CI-CD.md).
Provider Comparison¶
Reference pricing for Hetzner Cloud tiers and competitors (early-2026 list prices, kept for the sizing decision). Staging runs on the 8 GB / 4 vCPU tier (~EUR 7.49), production on a CX43 (16 GB / 8 vCPU).
| Provider | Tier | vCPU | RAM | Storage | Traffic | Price/mo |
|---|---|---|---|---|---|---|
| Hetzner | ~2 GB | 2 | 2 GB | 40 GB NVMe | 20 TB | ~EUR 3.79 |
| Hetzner | ~4 GB | 2 | 4 GB | 40 GB NVMe | 20 TB | ~EUR 4.99 |
| Hetzner | 8 GB (staging) | 4 | 8 GB | 80 GB NVMe | 20 TB | ~EUR 7.49 |
| Hetzner | 16 GB, CX43 (production) | 8 | 16 GB | 160 GB NVMe | 20 TB | ~EUR 14.99 |
| DigitalOcean | Basic shared | 2 | 4 GB | 80 GB SSD | 4 TB | ~EUR 22 |
| Contabo | Cloud VPS S | 4 | 8 GB | 75 GB NVMe | Unlimited | EUR 4.99 |
Prices as of early 2026. DigitalOcean price estimated from USD 24/month.
Why Hetzner¶
Technical Fit¶
PostgreSQL is tuned per environment. The staging configuration
(docker-compose.staging.yml) was tuned for a ~2 GB box and still runs with those values on
the 8 GB VPS:
shared_buffers=512MB,effective_cache_size=1536MBwork_mem=4MB,maintenance_work_mem=128MBmax_connections=50
The production configuration (docker-compose.prod.yml) was written for ~4 GB (container
mem_limits sum to roughly 3.75 GB) and runs on the 16 GB CX43 with plenty of headroom:
shared_buffers=1GB,effective_cache_size=3GBwork_mem=8MB,maintenance_work_mem=256MBmax_connections=50
The remaining memory covers Django/Gunicorn workers, Next.js, Redis, Celery, Nginx, and Docker overhead.
20 TB/month traffic far exceeds expected demand for a custom apparel e-commerce site.
NVMe storage is sufficient because media files live on DigitalOcean Spaces and database backups go to a separate Spaces bucket. The VPS only stores application code, Docker images, and the PostgreSQL data directory.
Cost Advantage¶
- Significantly cheaper than DigitalOcean for equivalent or better specs
- No hidden fees: IPv4 address included, firewall free, snapshots inexpensive
- Contabo offers similar specs, but Hetzner has a stronger track record for performance consistency and customer support
EU Compliance¶
- German datacenter locations
- ISO 27001 certified
- Meets GDPR data residency requirements for EU-based customers
Full TCO Breakdown (Monthly)¶
| Service | Provider | Cost |
|---|---|---|
| VPS staging (8 GB / 4 vCPU) | Hetzner Cloud | ~EUR 7.49 |
| VPS production (CX43, 16 GB / 8 vCPU) | Hetzner Cloud | ~EUR 14.99 (tier list price) |
| Volume (20 GB) | Hetzner | EUR 1.04 |
| Snapshot backups (10 GB est.) | Hetzner | EUR 0.12 |
| Domain (freezedesign.nl) | Yearly / 12 | ~EUR 0.75 |
| SSL certificates | Let's Encrypt | FREE |
| Monitoring | Sentry and PostHog free tiers (no external uptime monitor yet) | FREE |
| CDN | none | — |
| Media storage | DigitalOcean Spaces | ~EUR 5.00 |
| Database backups and media copy | DigitalOcean Spaces | included in the Spaces subscription |
| Total | ~EUR 29-30/month |
Annual cost: approximately EUR 350/year. Volume and snapshot lines are estimates; check the Hetzner invoice for the real amounts.
Pricing Notes¶
- Hetzner volumes: EUR 0.052/GB/month (20 GB = EUR 1.04)
- Hetzner snapshots: EUR 0.012/GB/month (10 GB = EUR 0.12)
- DigitalOcean Spaces: 250 GB storage + 1 TB transfer included for USD 5/month
- Database backups share the Spaces subscription with media (250 GB included); an hourly dump of an empty-catalog database is about 40 KB
Database Strategy¶
Decision: PostgreSQL in Docker container (not managed).
Why Docker PostgreSQL¶
- Hetzner does not offer managed PostgreSQL. Keeping the database on the same VPS avoids cross-provider latency and simplifies the architecture.
- Already proven. The application was developed and tested with Docker PostgreSQL.
Production tuning parameters are baked into
docker-compose.prod.yml. - Automated backups. Hourly
pg_dumpto DigitalOcean Spaces with tiered retention, AES-256-GCM encrypted. Scheduled by Celery Beat (backend/config/celery.py) — see Disaster Recovery. - No additional cost. Managed database services start at USD 15/month, which would more than double the TCO.
Configuration¶
- Image:
postgres:15-alpine(official PostgreSQL Docker image) - SSL/TLS: Enabled with self-signed certificates (
certs/server.crt,server.key) for container-to-container encryption - Backups: Automated hourly dumps to DigitalOcean Spaces (see Disaster Recovery); production deploys run
safe_migrate --dry-runfirst, which takes a backup before a destructive migration - Health checks:
pg_isreadywith 30-second intervals;max_connections=50
Managed Database Alternative¶
If future requirements demand managed PostgreSQL (automated failover, compliance mandates, database size exceeding VPS capacity):
| Provider | Plan | Starting Price |
|---|---|---|
| DigitalOcean Managed Database | 1 GB / 1 vCPU | ~USD 15/month |
| AWS RDS for PostgreSQL | db.t4g.micro | ~USD 15-20/month |
Migration path: standard pg_dump / pg_restore, then update DATABASE_URL in the
application environment.
Scaling Options¶
Vertical: Upgrade VPS Plan¶
The simplest scaling path. Hetzner allows live plan upgrades with brief downtime (5-10 minutes for VPS resize). No code changes required.
| Tier | vCPU | RAM | Storage | Price/mo |
|---|---|---|---|---|
| ~2 GB | 2 | 2 GB | 40 GB | ~EUR 3.79 |
| ~4 GB | 2 | 4 GB | 40 GB | ~EUR 4.99 |
| ~8 GB (staging) | 4 | 8 GB | 80 GB | ~EUR 7.49 |
| ~16 GB (production, CX43) | 8 | 16 GB | 160 GB | ~EUR 14.99 |
| ~32 GB | 16 | 32 GB | 320 GB | ~EUR 29.99 |
After upgrading, adjust the PostgreSQL tuning parameters and mem_limits in
docker-compose.prod.yml to take advantage of the additional RAM; the current values were
written for a 4 GB machine.
Managed Database¶
Offload PostgreSQL to a managed service while keeping the application on Hetzner:
- Migrate to DigitalOcean Managed PostgreSQL or AWS RDS
- Update
DATABASE_URLenvironment variable - Estimated total: EUR 7.49 (Hetzner) + ~EUR 14 (managed DB) = ~EUR 22/month
Consider this when:
- Automated failover and high availability become requirements
- The database outgrows the VPS disk or memory
- Compliance mandates separation of compute and data
Horizontal: Multiple VPS Instances¶
For significant traffic growth:
- Hetzner Load Balancer (EUR 5.49/month) distributing across multiple application VPS instances
- Managed or dedicated database server
- Redis Cluster or managed Redis for shared session/cache state
- A CDN (the Cloudflare design in Cloudflare CDN) for static assets; today nginx serves them with long cache headers
This level of scaling is unlikely to be needed in the foreseeable future for a custom apparel e-commerce site.
Why DigitalOcean Spaces for Media¶
Media storage uses DigitalOcean Spaces rather than Hetzner Object Storage:
- Already configured. The Django storage backend (
django-storagesS3Storage, enabled viaUSE_SPACES/DO_SPACES_*env vars) runs on staging and production. ams3(Amsterdam) region. EU datacenter, consistent with GDPR requirements.- Simple pricing. Flat USD 5/month for 250 GB storage and 1 TB outbound transfer. No per-request fees to worry about.
- Direct serving.
MEDIA_URLpoints at the bucket's own domain (<bucket>.ams3.digitaloceanspaces.com/media/), so product images never pass through the VPS; nginx's/media/location only serves the local media volume.
Switching to Hetzner Object Storage would save approximately EUR 1-2/month but would require reconfiguring the storage backend and re-uploading all media files. Not worth the effort at current scale.