There's common advice that anyone trying to learn Linux, DevOps, or SRE should build a homelab with as many physical nodes as possible. However, hardware is expensive, and I already have a single-node machine running my services. Buying several additional mini PCs or thin clients feels unnecessary when I could run multiple Linux VMs on my existing computer.
Would a VM-based lab let me learn most of the same skills? I'm especially interested in running a multi-node Kubernetes cluster with Talos, building a NAS, experimenting with Ceph, configuring pfSense, and learning Linux networking and lower-level system behavior. I'm not primarily trying to learn Terraform or Ansible, and I'm not looking for the easiest way to launch Kubernetes—I want to understand what happens when these systems are deployed more like bare-metal nodes.
What would I miss by using VMs instead of separate physical machines? Can I still realistically practice VLANs, routing, node failures, high availability, and recovery by treating VMs as disposable cluster nodes?
2 Answers
A VM lab can teach you a lot about high availability. You can stop a node, partition it from the network, corrupt a service, or simulate resource exhaustion and observe how Kubernetes or another distributed system reacts. You can practice quorum, failover, backups, and recovery without buying several machines.
The limitation is that all guests may still depend on one physical host. If that host loses power or fails, every node disappears at once. You also won’t learn much about failure domains unless you spread the VMs across multiple hosts. Keep that distinction in mind when evaluating an HA design.
A small dedicated host running Proxmox can be useful if you want the lab to stay online while you use your main computer, and it gives you a stable place to build VM templates and experiment with repeatable deployments. Terraform and Ansible are good additions, but they shouldn’t replace learning the underlying Linux and networking concepts.
That said, a dedicated host isn’t technically required. Running VirtualBox, libvirt, or another hypervisor on your existing Linux machine can provide nearly the same guest-level learning experience. Buy extra hardware later when you have a specific goal—such as testing physical networking, storage devices, power failure behavior, or multiple failure domains.
That makes sense. I can simulate individual node failures with VMs, but a single host remains a shared failure domain. I’ll treat that as a limitation rather than pretending it provides physical redundancy.