Kubernetes on bare metal and Proxmox

Running Kubernetes on your own hardware is cheaper at scale and keeps data in the building. It also means you own the pieces a cloud provider hides: load balancers, persistent storage, the control plane and its backups.

We build on-prem clusters with RKE2 or Talos for a hardened, reproducible base, MetalLB for service IPs, Longhorn or Ceph for persistent volumes and a proper ingress layer with cert-manager. Nodes are usually Proxmox VMs for easy rebuilds, or bare metal when GPUs or raw I/O matter.

At a glance

What RackLedge does

• Hardware and network design for HA
• Cluster bootstrap and hardening (CIS benchmarks)
• Storage class design and backups
• Integration with existing VLANs, DNS and identity

How we work

Every cluster we build starts from Git. Infrastructure as code creates the cluster, Flux or Argo CD reconciles everything inside it, and no change reaches production without a pull request. That is what makes a rebuild after a bad day a matter of hours instead of weeks.

Operations are the product: monitoring with alerts that mean something, backups of cluster state and persistent volumes, upgrades on a quarterly cadence and an on-call engineer who already knows the environment. We run production Kubernetes for our own platforms, so the runbooks are ones we use ourselves.

We are honest about fit. A handful of stable services on a couple of VMs does not need Kubernetes, and we will say so. Teams that ship weekly, need to scale or run many services get real value from it.

Related services

More on kubernetes

Frequently asked questions