Apache CloudStack — baking a PI

There is a rack on the bench in front of me: five Raspberry Pi 5 boards, sixteen gigabytes a piece, bolted into a stack the size of a hardback book, fed power and packets over a single PoE lead each. The whole thing draws less than a kettle. By the final chapter it is a genuine, working private cloud — and every layer of it is yours.

Most cloud computing happens somewhere you will never see, on hardware you will never touch, behind an invoice you cannot question. This book takes the opposite path. Over sixteen chapters, four of those boards become a real Apache CloudStack 4.22 cloud on ARM64 — management plane, NFS shared storage, two KVM compute modules, system VMs, guest networks, the lot — while the fifth stands apart as a local AI host carrying a Hailo-10H accelerator, an n8n orchestration engine, and an MQTT bus for ESP32 edge devices. By the end, the lab can sense its surroundings, decide what to do, act on the physical world, watch its own health, and rebuild itself from a Git commit.

What you build, in order:

• The foundations — an honest bill of materials, NVMe imaging and boot, a properly segmented VLAN design, and hosts prepared as code with Ansible
• The cloud itself — CloudStack 4.22 and MySQL on one board, NFS storage, the manual ARM64 system VM template (a step x86 users never see — you'll do it yourself and understand it), compute modules joined, and your first aarch64 guests behind a virtual router
• Infrastructure as code — the whole resource layer declared with OpenTofu, secrets in OpenBao, proved with a destroy-and-reapply round trip
• The edge and the AI — n8n workflows, MQTT and ESP32 sensors with a device-side safety interlock, and the silicon fork made explicit: which models run accelerated on the Hailo-10H and which run honestly, slowly, on the CPU
• Making it earn its keep — Prometheus/Grafana monitoring, the power and thermal budget, and ten tiered real-world uses from a whole-house DNS sinkhole to a GitOps self-service cloud where a commit provisions a VM.

The house style is honest framing. This lab is not highly available, not an x86 cloud, and not a benchmark champion — and the text says so every time it matters, because the gap between a working lab and a production cloud is mostly made of redundancy, and pretending otherwise would be the worst thing this book could do. What it is: one of the most honest ways to learn CloudStack there is, at one-hundredth the scale of an enterprise estate and none of the licence cost. Own it down to the last packet and the last watt.