After weeks of research and configuration, I have been running a complete self-hosted smart home stack on a Raspberry Pi 4 for over 7 weeks. Here is everything that is running and the lessons learned along the way.

The Stack

All services run in Docker on a single RPi 4 (4GB RAM) with a 500GB USB SSD.

Service	Purpose
Home Assistant	Smart home hub – 40+ devices
Frigate NVR	AI camera detection, no cloud
ESPHome	DIY ESP32 sensor management
Mosquitto	MQTT broker
Prometheus	Metrics collection
Node Exporter	RPi system metrics
Grafana	Monitoring dashboards
Speedtest Exporter	Internet speed tracking
Cloudflare Tunnel	Remote access, no port forwarding
Immich	Self-hosted photo backup
BookStack	Documentation wiki

Why Docker Instead of Home Assistant OS

The official recommendation is to use Home Assistant OS (HAOS) which gives you the supervisor and add-on store. That is a valid approach.

I chose Docker because I wanted everything in a single docker-compose.yml file – easier to backup, update, and migrate. The trade-off is that you configure everything manually, without the add-on store.

Key Lessons Learned

Use a USB SSD, Not a MicroSD Card

Running Docker volumes on a microSD card causes premature failure within months due to constant read/write. A 120GB USB SSD costs around €15–20 and will last for years. This is not optional for a production setup.

Pin Your Docker Image Versions

The grafana:latest tag pulled Grafana 12, which causes 100%+ CPU usage on ARM/RPi due to a new internal API server. Always pin to tested versions:

image: grafana/grafana:10.4.3  # not latest

Cloudflare Tunnel Over Port Forwarding

Remote access without exposing your home IP or opening ports on your router. Works even behind CGNAT. Free. Takes 15 minutes to set up.

cAdvisor on a Loaded RPi

cAdvisor for Docker container monitoring consumed 115% CPU on my RPi with 15+ containers, even with aggressive settings. Load dropped from 7.0 to 1.0 after removing it. For container monitoring on a loaded RPi, skip cAdvisor and use Home Assistant rest sensors for uptime monitoring instead.

Resource Usage

With the full stack running:

• CPU: 5–15% idle
• RAM: ~2.0 GB used of 3.7 GB
• Load average: 0.8–1.5

Getting Started

If you want to skip the research phase and deploy this stack with everything pre-configured, I put together a complete package with the docker-compose.yml, all config files, and a step-by-step setup guide:

👉 Home Assistant Complete Stack on Gumroad

It includes the full stack deployment guide (PDF), pre-configured config files for Frigate, Prometheus, Mosquitto, and ESPHome templates for adding your own sensors.

Questions? Feel free to reach out via the contact page or find me on Reddit as u/damian2609.