Homelab

The Goal: A Reproducible Server The worst kind of infrastructure is a server that nobody fully understands anymore. Things were installed over time, configuration changed in place, and the mental model of how it all fits together lives only in the memory of whoever last touched it. If that server dies, rebuilding it requires archaeology. The goal for mve-itguard was the opposite: a fresh Debian machine should go from zero to fully operational — users created, Docker installed, runners running, services ready to deploy — with a single command that anyone with the repository and the AGE key can run. ...

The Problem Every Project Has Every project that deploys to a real server has secrets. Tokens, passwords, API keys — values that should not be publicly visible but that the application needs to run. The GitOps approach makes this harder to ignore. If Git is the single source of truth and all configuration lives in the repository, where do the secrets go? Committing them in plaintext is an obvious mistake. Leaving them outside Git means the repository is no longer the complete source of truth — some knowledge lives only in someone’s head or in a separate system. ...

GitOps Is a Principle, Not a Tool When people talk about GitOps, Kubernetes quickly enters the conversation. FluxCD, ArgoCD, Helm — the ecosystem is rich, and the tooling handles a lot of the heavy lifting. The platform continuously watches the repository and reconciles the cluster state to match what Git declares. But the underlying principle has nothing to do with Kubernetes specifically. GitOps is simpler than that: Git is the single source of truth. No change happens outside of Git. Automation applies what Git declares. ...

The Problem with Commercial Solutions When I started thinking about home security cameras, the obvious path was a commercial solution — Ring, Nest, or one of the dozens of cloud-connected systems on the market. Plug it in, create an account, and you’re done. I did not go that route. Not because it is difficult to set up, but because of what it implies: your video footage lives on someone else’s server, your access depends on a subscription, and if the company changes its pricing or shuts down, you lose everything you built around it. ...

KubeCraft Course After completing the first Kubernetes course, where I deployed a basic K3s cluster on Rancher Desktop, I moved to the next level: a homelab course focused on building a fully GitOps-based K3s cluster on a Debian VM, hosting real self-hosted applications exposed to the internet. What to Expect at the End? The final goal is to deploy a self-hosted Linkding application accessible from the internet using a custom domain name. ...

My First Homelab (for the second attempt) This homelab is the first real infrastructure I’m building—although technically, it’s already my second attempt. Why a Second Attempt? The first time I tried to set everything up, I used Proxmox running on a VM provider and Talos Linux for my Kubernetes nodes. I experimented with countless configurations, trying to understand how Talos works behind the scenes. Each time a node failed or something didn’t behave the way I expected, I had to rebuild everything manually, one node at a time. It was painful, slow, and sometimes discouraging… but I learned a lot from it. ...

Learn persistent volume on kubernetes My status in my learning session on kubernetes : Name Learning stat Commentary Pods Create, delete, logs Deployment 🟢 Consepts understood nfs-driver 🟢 In place to use TrueNAS NFS shared storage Calico 🟠 Basic functionality; not fully mastered but operational Service 🟠 ClusterIP and NodePort understood; LoadBalancer and ExternalName not yet used Traefix 🟠 Learning, used with Service LoadBalancer API gateway 🟠 Learning, used with Service LoadBalancer PV 🔴 Persistent volumes not yet required PVC 🔴 Persistent volume claims not yet required I hadn’t needed to create a PV and a PVC. So, I had to learn how to access this ressource in my Talos Kubernetes cluster. ...

Why I run OpenClaw at home I want an AI agent who can read my files, run commands, and help me on my homelab project, but I don’t want to send my workspace to a SaaS or open a public port on my router. OpenClaw is a self-hosted agent platform: it run on my Linux box, hold the agent sessions, and expose a small control UI to interact with it from any of my devices. ...

Building a Home Lab to Learn Kubernetes I wanted to learn Kubernetes—really learn it, not just follow a few tutorials. While researching, I noticed many people build their own home labs to run cloud services or custom applications. That inspired me to set up a complete home lab myself: first to run basic services like Nextcloud or OpenProject, and then to experiment with a full DevSecOps pipeline. First Step To understand Kubernetes properly, I needed something close to real-world infrastructure. Buying three or more physical servers wasn’t possible, so I decided to create everything on a single machine using multiple virtual machines (VMs). This way I can spin up as many servers as I need and quickly learn how to build a cluster. ...

First issue Multiples environnements First, I need to connect to my server in SSH and in http to manage Proxmox. With this diagram you can see a simple usage of VM machine. But, I have personal conditions. I need to have a virtual network to have a stable IP of my node in my cluster. But, as you can see, in default with proxmox, you have one bridge created connect to the physical Ethernet port in the server. And, I change environnement to work in my office, and my home because I don’t have an internet box in there, so for connect my server to my PC, I need to connect the Ethernet ports of the server and my PC in direct. ...