Skip to main content

Requirements

Prerequisites

Two hosts cannot have the same name. By default, the node name is taken from the machine's hostname.

Hardware

Hardware requirements scale based on the size of your deployments. Minimum recommendations are outlined here.

Server CPUServer RAMNumber of Agents
816 GBsmall
16+32 GBstandard

Disks

SSD when possible to ensure optimal performance (at least for the cluster controllers)

Networking

Inbound Network Rules

PortProtocolSourceDestinationDescription
443TCPWorker nodesWorker nodesIngress API
6443TCPWorker nodesController nodesKubernetes API
9345TCPWorker nodesController nodesKubernetes supervisor API
10250TCPAll nodesAll nodeskubelet metrics
2379TCPController nodesController nodesetcd client port
2380TCPController nodesController nodesetcd peer port
2381TCPController nodesController nodesetcd metrics port
30000-32767TCPAll nodesAll nodesNodePort port range
8/0ICMPAll nodesAll nodesCilium CNI health checks
4240TCPAll nodesAll nodesCilium CNI health checks
51871UDPAll nodesAll nodesCilium CNI with WireGuard

OS

If possible, you should use the hardened OS from the delivery which is based on a base image ubuntu 22.04.5 in kernel 119 available https://releases.ubuntu.com/22.04/ubuntu-22.04.5-live-server-amd64.iso . The GPU part is sensitive to the kernel version so if ever these prerequisites are not respected then you will have to bring the necessary image for the gpu-operator. By default, the installation images of the nvidia drivers in version 535 and compatible with kernel 119 are available in the delivery. The GPUs must be visible at the OS level (if vm, consider "passthrough")