Production homelab: 50+ services on 3 nodes

Three Ubuntu boxes. One Caddy reverse proxy. Fifty-plus self-hosted services running the show — email, git, CI/CD, dashboards, password manager, photo backup, and a small Kubernetes cluster for the things that need it.

The stack

Node	Role	Specs
`edge-01`	Reverse proxy + firewall	2 vCPU, 4GB RAM, public IP
`app-01`	Application workloads	8 vCPU, 32GB RAM, NVMe
`data-01`	Databases + storage	8 vCPU, 32GB RAM, 2×4TB HDD (RAID1)

What runs where

edge-01 (public-facing):

Caddy with auto HTTPS via Cloudflare DNS-01
Crowdsec + fail2ban for intrusion detection
WireGuard for admin access

app-01 (workloads):

Woodpecker CI
Gitea (lightweight self-hosted Git)
Drone runners (3 instances)
Plausible Analytics
Uptime Kuma
Healthchecks.io (cron monitoring)

data-01 (stateful):

PostgreSQL 16 (single primary, streaming replication to app-01)
Redis 7
MinIO (S3-compatible, for backups + media)
Restic repo (nightly snapshots → B2 + local)

Orchestration

Single docker-compose.yml per service, kept in a private Gitea repo. Deploys via a small bash script that:

Pulls latest config from Gitea
Runs docker compose pull
Health-checks before stopping the old container
Rolls back automatically if health check fails

Zero-downtime deploys for everything that’s stateless.

Monitoring

Prometheus on data-01, scraping all 3 nodes via node-exporter
Grafana dashboards for CPU, RAM, disk, network, Docker containers
Uptime Kuma for external service health
Alertmanager → Telegram when anything’s off

Backups

Restic → Backblaze B2 (encrypted, deduplicated)
Hourly snapshots of databases, daily of everything else
30-day retention + monthly archive
Monthly restore drill (yes, I actually test)

What’s next

Replace single PostgreSQL with Patroni + etcd (HA)
Move stateless workloads to a small k3s cluster
Add Ceph for proper distributed storage

Full config and dashboards: github.com/nuteo/homelab (placeholder)