hyggecloud / deployment
Deployment you can actually reason about
No click history in a web console, no server only known to the colleague who's currently on holiday. Every environment we build is infrastructure as code — versioned, documented, reproducible. That's not ideology, it's your exit insurance: including from us.
Principles
Three rules for every environment we build
git clone instead of wiki archaeology
Terraform for infrastructure, Ansible for configuration, Helm/Kustomize for workloads. The entire environment lives in your repository — every change is a commit with an author, a date and a reason.
The cluster follows the repo
Deployments run through the pipeline, not through SSH sessions at 11pm. What's in main runs in production — and what runs in production is in main. Rollback = git revert.
Boring is the new robust
We build on tools with ten years of future: PostgreSQL, Kubernetes/k3s, Nginx/Traefik, Debian. No experiments at your expense — innovation belongs in the product, not the infrastructure.
Reference architectures
Three sizes. No overengineering.
The most common post-migration disease: architectures built for Google scale, serving three customers. We size for reality — and build so that growth is an upgrade, not a rebuild.
S — The solid start
2–3 VMs + Docker Compose. Reverse proxy (Traefik/Caddy) with auto-TLS, app containers, PostgreSQL with daily backups, monitoring agent.
- For MVPs, internal tools, setups up to ~10k users
- From ~€50/month infrastructure cost
- Deploy via CI pipeline in < 2 minutes
- Upgrade path to M without re-architecting
M — The growing product
k3s cluster (3+ nodes) with load balancer, cert-manager, GitOps (Argo CD/Flux), central logging and HA Postgres (Patroni or managed).
- For SaaS products with real traffic and SLOs
- Rolling deployments without downtime
- Horizontally scalable by adding nodes
- Typically €200–800/month instead of a four-figure AWS bill
L — The HA setup
Managed K8s or a multi-node cluster across two locations, Ceph/Longhorn storage, object storage, dedicated monitoring, defined RPO/RTO.
- For platforms with availability commitments
- Site failover (e.g. Nuremberg ↔ Helsinki)
- Disaster recovery plan incl. restore tests
- Details on the backups page
What it looks like
A HyggeCloud deployment — from commit to live
# CI: tests .............................. ✓ 214 passed
# CI: container build .................... ✓ app:v2.4.1
# CI: security scan (trivy) .............. ✓ 0 critical
# CD: rolling update (k3s) ............... ✓ 6/6 pods ready
# Health checks .......................... ✓ 200 OK
➜ Deploy duration: 94 seconds · downtime: 0
➜ Rollback if needed: git revert && push
Always included
- Automatic TLS — Let's Encrypt via cert-manager or Traefik, never an expired certificate
- Secrets management — sops/age or OpenBao, no passwords in repos or pipelines
- Staging environment — identical to production, because "works on my machine" is not a deployment strategy
- Monitoring from day one — Grafana, Prometheus, Loki and alerts before the first user arrives
- Documented runbooks — deployment, rollback, incidents: everything written down, nothing in one person's head
// The toolkit
Straight talk
What we do not do in deployments
- No Kubernetes at any cost. For many setups, Docker Compose on two solid VMs is the better answer. You get K8s when you need K8s.
- No tool zoos. Every extra tool is a tool your team has to learn and maintain. We pick few, proven components.
- No lock-in through the back door. Our own setups are built so that any competent admin can take them over.
"The best infrastructure is the one nobody has to talk about.
It deploys on Mondays like on Fridays, it survives a node failure, and when someone asks how it works, you point at a repository instead of a person."
— The HyggeCloud principle
Show us your setup — we'll show you the path.
In the Hygge Check we analyse your current deployment landscape and design the target architecture — sized for reality, not for glossy diagrams.
→ Book an intro call30 minutes · no strings attached · engineers, not salespeople