Hardware¶
Everything lives in a 12U rack.
Cluster nodes¶
Three Talos nodes. All three are control plane nodes and also run workloads.
| Node | System | CPU | RAM | Node IP |
|---|---|---|---|---|
k8s-01 |
Lenovo ThinkCentre M920x | i9-9900T | 64 GB | 10.73.20.10 |
k8s-02 |
Lenovo ThinkCentre M920x | i9-9900T | 64 GB | 10.73.20.20 |
k8s-03 |
Lenovo ThinkCentre M90q Gen 5 | i5-13400T | 64 GB | 10.73.20.30 |
Common to every node:
| Part | Model | Use |
|---|---|---|
| OS disk | Kingston NV3 1 TB (NVMe) | Talos install, EPHEMERAL (capped at 160 GiB), local-hostpath user volume |
| Ceph disk | Micron 7300 PRO 480 GB (NVMe) | One Rook-Ceph OSD per node |
| NIC | Intel X520-DA2 (10G SFP+) | bond0 (active-backup), aliased as net0 by MAC |
| iGPU | Intel UHD | Hardware transcoding via the Intel GPU resource driver (DRA) |
| Out-of-band | JetKVM + DC Power Control Extension | Remote console and power |
Finding a node's hardware details
Each node file in
kubernetes/talos/nodes/controlplane/
starts with a comment block listing its model, NIC MACs and drivers, and disk
IDs. Disks are selected by model and links by MAC, not by kernel name, so
nvme0/nvme1 swapping across reboots is harmless.
NAS¶
TrueNAS SCALE on a self-built 3U server.
| Part | Details |
|---|---|
| CPU / RAM | Intel i7-6700K, 64 GB |
| Boot | WD Red SA500 500 GB (SATA SSD) |
| Bulk pool | 6 × 14 TB Toshiba MG09 (SATA), 1 × 6-wide RAIDZ2 |
| 5 × 4 TB HGST Ultrastar 7K4000 (SAS), 1 × 5-wide RAIDZ1 | |
| Fast pool | 2 × 1 TB Crucial MX500 (SSD), mirrored |
| NIC | Intel X520-DA2 (10G SFP+) |
| Out-of-band | JetKVM + ATX Extension Board |
Network¶
| Device | Role |
|---|---|
| UniFi UDM Pro Max | Router, firewall, BGP peer, DNS, NVR (1 × 8 TB Seagate SkyHawk AI) |
| UniFi USW Aggregation | 10G SFP+ aggregation switch; cluster nodes and NAS |
| UniFi USW Pro HD 24 PoE | 2.5G/10G PoE++ core switch |
Power¶
| Device | Role |
|---|---|
| UniFi UPS 2U | 1500 VA rackmount UPS, monitored by nut-exporter |