
Proxmox Backup Lab
DiscoverA Proxmox hypervisor administered entirely with Ansible. I set up the services, back them up and then cause a disaster on purpose to prove the restore actually works.
Yosef Naabid
I run the production systems of a European hosting provider. I test what I administer by breaking it before I trust it, and a backup only counts once I've restored it.

I'm the only one who answers for the systems at Meridia Softworks, a hosting provider with clients across Europe. Its infrastructure is written as code and rebuilds from scratch in 15 minutes, which is why I could move the whole thing to a dedicated server without a single client noticing. The 20-odd services run on the same machine, each sealed in its own container, so the one that goes haywire can't drag the rest down. Administration is no longer reachable from the internet, only through a WireGuard tunnel that terminates on the router itself. If something goes down, the alert fires in under a minute and the service brings itself back up.

PgO UCAM is 90 sites across 27 countries, the Catholic University of Murcia's network of dental postgraduate programs. I migrated 20 physical servers to 25 virtual machines, between the data center's VMware cluster and Proxmox hosts at the sites, and a failed host stopped meaning a failed service. I split the network into 6 zones for more than 150 devices, so there were always two barriers between an infected machine and the clinical records. Those backups I left tested for the GDPR audit, in versions ransomware cannot touch. The monitoring for the 10 clinics I built with Zabbix and Grafana, and an outage stopped being something a user reported by phone. Along the way I cut more than €5,000 a year in Microsoft 365 licensing.

Clínicas DEA is the clinical side of the same group as PgO UCAM, so I ran its infrastructure alongside. The clinic workstations and accounts hung off the same Active Directory and the same Microsoft 365, and its servers, holding patient data under GDPR, were part of the backup plan with a restore verified every month. Day-to-day L2 support went through me as well.

A Proxmox hypervisor administered entirely with Ansible. I set up the services, back them up and then cause a disaster on purpose to prove the restore actually works.
Six VLANs behind an OPNsense where the allowed flows are declared in one file. I don't write the deny rules, the code computes them by subtracting what's allowed from every possible combination, and a suite tests, one by one, the 266 that must fail.
Monitoring as code. I bring up a complete Zabbix with one command, and hosts and alerts are defined through the API instead of by clicking around.

A complete Windows domain brought up and administered from PowerShell instead of the GUI. Repeatable, auditable and hardened in line with the CIS Benchmark.

A three-node Kubernetes cluster I bring up from code, govern from Git and break on purpose with four timed drills. The last one destroys all three nodes, and everything comes back, data included, in under ten minutes.

The site you're looking at runs on an Azure VM that I built and administer myself. AlmaLinux, Nginx, HTTPS with Let's Encrypt and Cloudflare in front, all configured by hand. If it goes down at 3 a.m., the one who SSHes in is me.