Building a Kubernetes cluster on Proxmox VMs
Three VMs on a single server are enough to learn — and enough for many real workloads.
Starting at from €13.99/mo
Introduction
Running Kubernetes on Proxmox virtual machines is the cheapest way to have a cluster of your own. A single VDS carries the three nodes of a learning cluster, and the hypervisor gives what Kubernetes does not: a snapshot before every risky move.
On sizing, tell two cases apart. A k3s cluster — the lightweight distribution — fits on three 4 GB VMs, so on the S plan at €13.99. A full Kubernetes asks for more: count 4 GB per control node and 8 GB per worker, which makes the XL plan at €37.99 a reasonable starting point.
One thing cannot be wished away: on a single server, a three-node cluster does not survive the server failing. Kubernetes redundancy protects against losing a node, not against losing the machine carrying them all. Real availability needs several servers — and that is another budget, to be taken on knowingly.
What the server brings, across the whole range
Proxmox VE already installed
The server arrives with <strong>Proxmox VE</strong> in place and its web interface reachable. You create your virtual machines and LXC containers from the browser, without a support ticket or a wait. <strong>Reinstalls are unlimited</strong>: an experiment that goes wrong is redone in minutes.
RAM, because that is what runs out first
On a hypervisor it is almost never the processor that caps out first: it is <strong>memory</strong>, because every virtual machine reserves its own. The range starts at <strong>16 GB</strong> and goes up to <strong>96 GB</strong>, which leaves room to stack several VMs without starving them.
Unlimited snapshots and backups
A snapshot before an update, another before a configuration change: that is what separates a test from a gamble. Neither the count nor the frequency is billed, so nothing discourages taking one more.
Up to 16 additional IPv4 addresses
Each VM can hold its own public address, which avoids routing everything through port forwards on the host. Useful as soon as a virtual firewall, a web server and a third-party service have to coexist cleanly.
What people actually do with it
Learn Kubernetes without going broke
Three VMs on a €13.99 server beat an hourly managed cluster when starting out. And unlimited reinstalls forgive mistakes.
A staging environment
Replay production manifests before applying them for real. The hypervisor snapshot doubles the safety net.
Prepare a certification
Kubernetes exams demand practice on a cluster you can break. This one resets in minutes.
Host small workloads
A three-node cluster on the XL plan runs real applications, provided you accept that a server failure is a total failure.
Test a version upgrade
Moving from one minor version to the next is exactly the kind of operation you rehearse somewhere other than production.
Frequently asked questions
k3s to learn and for small workloads: it installs in one command and uses little. Full Kubernetes earns its place when you must faithfully reproduce an existing production environment.
Three 4 GB nodes fit on the S plan, but without headroom. The L plan at 32 GB or XL at 48 GB gives room to work and to add workers.
No, not on a single server. Kubernetes will reschedule a pod whose node dies, but if the physical server stops, all three nodes stop together. Several servers are needed for redundancy to mean anything.
On a single server, a local provisioner is enough and stays simple. Longhorn or Ceph become relevant when you spread across several servers — and then cost memory and network.