Unit 4: Our real-world case - Hybrid model

We are going to show you how our real infrastructure works. It is not theory, these are not made-up numbers. This is what we have set up right now, providing a service every single day.

pve1 - The local server (in the office)

• What it is: a physical server installed in our office. Together with pve2 it forms a two-node Proxmox cluster
• Processor: Intel Xeon E5-2697v3 (14 cores, 28 threads) - a professional server processor
• RAM: 62 GB - enough to run 15 applications at once without breaking a sweat
• Storage: NVMe of roughly 4 TB with the ZFS file system - ultra-fast disks with protection against data corruption
• System: Proxmox 9.2.2 - a virtualisation platform that lets you create multiple "virtual computers" inside a single physical one
• Active containers: about 15, each with its own isolated application
• Network: 192.168.1.2 - accessible only from the office local network
• What we use it for: software development, testing, artificial intelligence tools (Claude Code), storage of sensitive data, backups

This server is not accessible from the internet. It sits behind the office door, literally. It is our "digital safe".

pve2 - The local AI testing node (in the office)

• What it is: a second physical server in the office. Together with pve1 it forms a two-node Proxmox cluster, which lets us spread the load and move machines from one node to the other when needed
• Processor: Intel Xeon E5-2680 v4 (14 cores, 28 threads) - another professional server processor
• RAM: 31 GB
• Graphics card: NVIDIA GeForce RTX 3060 OC with 12 GB of VRAM - the heart of this node, the one that accelerates the artificial intelligence models
• Storage: a 4 TB hard drive for data and backups, plus two NVMe SSDs (of 512 and 500 GB) for the system and the machines that need speed
• System: Proxmox 9.2.2, the same version as pve1 (the two nodes of the cluster stay in sync)
• Active containers: 3, dedicated to artificial intelligence testing
• Network: accessible only from the office local network, just like pve1
• What we use it for: it is our local AI test bench. We run artificial intelligence models on our own premises, without sending data to external services, to research and learn before taking anything into production

Having pve1 and pve2 in a cluster gives us room to grow: if one node runs short, the other takes on part of the work, and we can experiment with AI without touching the services that are already running.

pbs - The backup server (in the office)

• What it is: a machine dedicated exclusively to storing backups. It runs Proxmox Backup Server (PBS), a software specialised in backing up virtual machines and containers
• Address: 192.168.1.4 within the local network, with its own disks organised into two independent stores
• What it backs up: the two nodes of the cluster (pve1 and pve2) and one or two more machines, each in its own separate space within the backup server
• Capacity: on the order of 4 TB per store; right now it has about a quarter of it used
• What it is for: it makes automatic backups every day, keeps only what has changed since the previous backup (deduplication: a daily backup takes up very little), and periodically verifies that those backups can genuinely be restored

The key to pbs: it keeps the backups on a separate machine, never on the same server as the original data. If pve1 or pve2 break down, their backups remain intact on the backup server. It is the lesson of the OVH fire applied at home: the backup and the original must never live together.

ipve1 - The remote server (at OVH)

• What it is: a dedicated server rented from OVH, in a data centre in France
• Processor: Intel Xeon E5-1650v2 (6 cores, 12 threads)
• RAM: 31 GB
• Storage: SSD of roughly 200 GB
• System: Proxmox 8.4, about 8 containers
• Network: publicly accessible as lemontreecloud.com (IP 54.36.179.223)
• What we use it for: production Odoo (the one customers use), email, public websites, remote access

This is the server that is "public-facing". Customers access it from anywhere with internet, 24 hours a day.

Why hybrid? The best of both worlds

The combination of both servers gives us advantages that no single model can offer:

Full control over sensitive data: the source code, the development tools and the backups are on the local server. Nobody accesses them except us, physically, in the office.

24/7 remote access for customers: the production Odoo is at OVH, accessible from any browser in the world. Customers do not need a VPN or special software.

Redundancy: if the local server has a problem, the production Odoo at OVH keeps running. And if OVH has an incident, we have local backups to restore from.

And the cost?: between 60 and 100 euros a month for the whole infrastructure. To put it in perspective: an equivalent setup on AWS would cost between 300 and 500 euros a month. And an Odoo SaaS for the same number of users, more than 250 euros a month in licences alone.

Sixty to a hundred euros a month. Less than the mobile phone bill of many self-employed people.