Skip to main content

Hetzner Bare Metal & HCloud Server Purchasing Guide

This document provides guidance for purchasing Hetzner servers for Kubernetes deployments, including recommended hardware, cluster layouts, and best practices.

1. Supported Cluster Layouts​

We support three types of cluster layouts:

LayoutDescription
All HCloudControl plane and worker nodes fully deployed on HCloud servers.
All Bare MetalControl plane and worker nodes fully deployed on bare metal servers.
HybridControl plane on HCloud servers and worker nodes on bare metal servers.

2. Bare Metal Server Recommendations​

Note: Bare metal clusters have no limits on volume creation but do not support automatic node scaling.

For bare metal clusters, we recommend:

  • Single server setup: One node (suitable for testing or small deployments).
  • High-availability setup: 3 control plane nodes + 3 worker nodes for redundancy. (we recommend running the control plane on VMs to save on resources and cost).

Best Practices​

  • Purchase servers via Hetzner auctions to reduce costs.
  • Ensure control plane nodes are located in the same country but across different datacenters to maintain redundancy.
  • Include separate backup servers for critical data.
  • We recommend control planes to be VMs (HCloud node) to save resources and cost.

Hardware Recommendations​

Backup Server​

  • Storage: 4 × 10 TB SATA Enterprise HDD
  • RAM: 4 × 8192 MB DDR3 ECC
  • Network: 1 Gb NIC

Control Plane Node​

  • Storage: 40 GB SSD
  • RAM: 4 GB
  • Notes: Control plane nodes will not run many workloads.
  • If you need to schedule workloads on control plane nodes, you can use the worker node configuration below; however, this is not recommended for security reasons. The pods running on the control plane node should not run public ingress services.

Single-Host Cluster / Worker Node​

  • Storage: 2 × 1 TB SSD M.2 NVMe
  • RAM: 4 × 16384 MB DDR4

Optional Networking Enhancement​

  • Add a 10 Gb NIC to bare metal nodes to boost internal connectivity (cost: ~5€ per NIC per node).
  • Only required for internal node traffic; a 10 Gb uplink is not necessary. A 10 Gb uplink can cost 43+€ and is only required if high-speed internet connectivity is needed.

3. HCloud Server Recommendations​

  • Control Plane Node: CAX11 (2 vCPU, 4 GB RAM, 40 GB storage, ~4€/month)
  • Worker Node: CAX41 (16 vCPU, 32 GB RAM, 320 GB storage, ~23€/month)

Important Notes​

  • HCloud nodes have a limit of 16 volumes per node.
  • Volume attachment is subject to location limitations per country, so plan accordingly.

4. Hybrid Cluster (Recommended setup)​

  • Control Plane: CAX11 HCloud server (~4€/month)

  • Worker Node: Bare Metal

    • RAM: 4 × 16384 MB DDR4
    • Storage: 2 × 1 TB SSD M.2 NVMe

Optional Networking​

  • Add a 10 Gb NIC to bare metal worker nodes to improve internal connectivity (~5€ per NIC).

5. Cluster Configurations Chart​

LayoutNodesStorageRAMNotes
Single Bare Metal1 node2 × 1 TB NVMe4 × 16 GBSmall/test cluster
HA Bare Metal3 control + 3 worker2 × 10 TB HDD4 × 16 GBProduction with redundancy
All HCloud3 control (CAX11) + 3 worker (CAX41)CAX41 defaultCAX41 defaultCloud cluster with scaling
Hybrid1 or 3 control (CAX11) + 1 or 3 worker (bare metal)2 × 1 TB NVMe4 × 16 GBInternal high-performance workloads

6. Summary of Best Practices​

  • Always distribute control plane nodes across datacenters.

  • Use Hetzner auctions for bare metal servers to reduce costs.

  • Match node specifications to their role:

    • Control plane: moderate RAM + SSD
    • Worker nodes: large storage + ECC RAM

  • For hybrid setups, HCloud provides flexibility and easy scaling, while bare metal delivers high performance.

  • Consider optional 10 Gb NICs to improve intra-cluster communication on bare metal nodes.