Servers

Overview

The homelab spans three physical locations: bare-metal at home, a Raspberry Pi, and an Oracle Cloud VPS — all connected via a Tailscale mesh VPN.

Server Inventory

Server	Role	Site	OS
k3s-m1	k3s master	🏠 Home	Ubuntu 22.04 LTS
k3s-m2	k3s master	🇵🇹 PT	Ubuntu 22.04 LTS
k3s-oci-m3	k3s master	☁️ Oracle Cloud	Ubuntu 22.04 LTS
rpi-4b	k3s worker + Docker host	🏠 Home	Ubuntu 24.04 LTS
rpi-z2w-hyperion	LED controller (Hyperion)	🏠 Home	Ubuntu 24.04 LTS
hoarder	NAS / storage	🏠 Home	Unraid v7.2.4

Ansible Inventory Groups

Defined in ansible/inventory/hosts.yaml:

masters:        # k3s-m1, k3s-m2, k3s-oci-m3  (k3s control plane)
nodes:          # (k3s agent nodes — currently empty)
k3s_cluster:    # masters + nodes
raspberries:    # rpi-4b, rpi-z2w-hyperion
docker:         # rpi-4b (also runs Docker directly)
unraid:         # hoarder
hyperion:       # rpi-z2w-hyperion

Oracle Cloud Server (k3s-oci-m3)

The OCI master runs on Oracle's Always Free tier and is connected to the cluster exclusively via Tailscale. This requires special DNS handling:

Custom resolv.conf required

Oracle VMs have a VCN search domain appended to /etc/resolv.conf. This causes transient DNS failures inside pods after server restarts.

Fix: create /etc/k3s-resolv.conf (without the Oracle search domain) and configure k3s to use it:

# /etc/rancher/k3s/config.yaml (on k3s-oci-m3)
kubelet-arg:
  - "resolv-conf=/etc/k3s-resolv.conf"

Tailscale Connectivity

All servers are connected via Tailscale. Server IPs on the Tailscale interface (tailscale0) are used for k3s inter-server communication, configured in ansible/inventory/group_vars/k3s_cluster.sops.yml:

server_config_yaml: |
  node-ip: "{{ ansible_facts['tailscale0']['ipv4']['address'] }}"

Tip

This means the cluster continues to work even if the home WAN IP changes, and OCI traffic never touches the public internet.