Skip to main content
Back to Blog

How I Self-Host Everything on a Home Server with 99.99% Uptime

8 months ago3 min read
devopsself-hostingdockertailscale

When most developers ship a product, they reach for AWS or Vercel. I built a home server instead — and it runs everything: Shorty, uNotes, Caramel, UpUp, this portfolio, and devino.ca. With 99.99% measured uptime.

Here's the full stack and why each piece was chosen.

The Hardware

A refurbished Dell PowerEdge R710 with 48GB RAM and dual Xeon processors. Cost: $200. Monthly electricity: ~$15. Compare that to AWS EC2 for equivalent compute: $300+/month.

Orchestration: Dokploy on Docker Swarm

Dokploy is an open-source Heroku/Vercel alternative you self-host. It gives you a UI for deploying apps, managing environment variables, and viewing logs — all on your own infrastructure.

# Deploy an app with nixpacks (auto-detects your stack)
dokploy deploy --app my-nextjs-app --nixpacks

Docker Swarm handles replication and service discovery. Each app gets its own container, and Swarm restarts failed containers automatically.

Networking: Tailscale + Cloudflared

The internet never touches my home network directly. Here's the flow:

User → Cloudflare CDN → cloudflared tunnel → Tailscale mesh → Home server

Cloudflared creates an outbound-only tunnel from my server to Cloudflare's edge. No port forwarding, no firewall holes. Cloudflare handles DDoS protection and TLS termination for free.

Tailscale creates a WireGuard mesh between all my devices. My laptop, phone, and servers all get private IPs (100.x.x.x) that can reach each other securely from anywhere.

Reverse Proxy: Traefik

Traefik auto-discovers Docker containers and configures routing automatically. Zero-downtime deploys just work.

labels:
  - "traefik.http.routers.app.rule=Host(`myapp.com`)"
  - "traefik.http.services.app.loadbalancer.server.port=3000"

Monitoring: Uptime Kuma + Grafana

Uptime Kuma pings every endpoint every 60 seconds and sends alerts via Telegram. It's how I know the 99.99% number is real, not aspirational.

Grafana + Prometheus gives me CPU, memory, and request metrics. All on the same home server.

Why Not Just Use the Cloud?

Cost: $215/month vs ~$15/month all-in.

Control: I own the data. No vendor lock-in. No surprise bills.

Learning: Running your own infra teaches you things no managed service does. I've debugged kernel OOM events, ZFS pool degradation, and Swarm leader elections. All valuable.

The downside is real: you need to be ready to SSH in at 2am when something breaks. For me, that's fine. For a business with SLA requirements, factor that in.

The setup isn't for everyone. But if you're a developer who wants to understand infrastructure from the ground up — and save a lot of money — a home server is worth every hour of setup time.

All posts