Can I run a small Kubernetes cluster on two Windows machines for students?

I’d start with Linux VMs rather than trying to make WSL2 the foundation. Kubernetes networking between separate WSL2 installs can take extra work because of NAT, and that can make a two-machine cluster awkward to connect and troubleshoot. A small Ubuntu VM on each machine gives you a more conventional Linux setup.

Keep expectations modest with 24 GB of RAM across both hosts. It may be fine for a handful of lightweight exercises, but student workloads can quickly compete for memory and CPU. Test the actual workloads and set conservative requests, limits, and quotas before inviting a class to use it.

For a personal lab, I’d try k3s inside Linux VMs. It’s a smaller Kubernetes distribution and can be a gentler way to learn the basics. I wouldn’t count on Windows itself as a supported k3s node, and I’d treat WSL2 as an experiment rather than the reliable path for a cluster spanning two computers.

Before building the student workflow, get one cluster running and check that nodes can reach each other, DNS works, and a test volume survives a pod restart. Those checks will tell you more than adding monitoring or per-student automation right away.

One cluster with a namespace per student is a useful starting point, but a namespace alone is not a strong security boundary. Add appropriate access controls and network policies, and be careful about what students are allowed to create. Resource quotas help cap consumption, but they don’t make a small cluster capable of serving unlimited workloads.

For storage, decide how you’ll provide a separate persistent volume for each student and what happens when the cluster or a host is unavailable. With only two machines, there’s limited room for resilience, so treat this as a learning setup rather than dependable classroom infrastructure.

If the goal is mainly to let students write code and use Kubernetes, consider giving each person a browser-based development workspace and a more isolated Kubernetes environment, instead of letting everyone share one cluster directly. That can be easier to control, though it adds setup and resource costs. There are tools that provide virtual clusters, but I’d first confirm that the hardware can handle the simpler version.