Platform Setup
This section explains the host layer beneath Kubernetes: how Talos Linux was installed, how the NVIDIA GPU became available to workloads, and how local persistent storage was prepared during the same installation.
Talos machine configuration is maintained separately from this repository. The separation is intentional:
flowchart TB
TalosConfig[Talos machine configuration] --> Host[Talos Linux node]
Host --> Kubernetes[Kubernetes]
Git[Homelab Git repository] --> Flux[Flux CD]
Flux --> Kubernetes
Kubernetes --> Workloads[Services and applications]
- Talos configuration owns the operating system, kernel modules, node scheduling policy, and host storage mounts.
- This repository owns Kubernetes resources reconciled by Flux.
- Documentation here connects the two layers without publishing Talos credentials or generated machine configuration.
Read in order
- Talos Linux follows the complete
v1.13.6single-node installation in execution order, including the NVMe storage layout. - NVIDIA GPU explains why a custom Talos image and kernel modules were required before the Kubernetes GPU Operator could work.
After Talos creates and mounts the XFS user volume, the
Local Path Provisioner exposes it to
Kubernetes workloads through the local-path StorageClass.
The host was rebuilt from one validated controlplane.yaml. The custom
installer, NVIDIA modules, control-plane schedulability, EPHEMERAL limit, and
XFS user volume were not applied as later patches.