🏗️ INFRASTRUCTURE
Fondation bare-metal
The bare-metal foundation: immutable OS, container runtime, and cluster orchestration.
TOPOLOGIE DU CLUSTER
Control Plane (3 nodes — HA with Raft consensus) ├── controlplane-100 192.168.0.100 Ready ├── controlplane-101 192.168.0.101 Ready └── controlplane-102 192.168.0.102 Ready API Server VIP (ARP-based failover) └── k8s.edgeprime.io 192.168.0.250 Active ~10s failover Worker Nodes (5 nodes — workload execution) ├── worker-103 192.168.0.103 Ready ├── worker-104 192.168.0.104 Ready ├── gpu-105 192.168.0.151 Ready 3× NVIDIA GPU ├── worker-106 192.168.0.150 Ready └── gx10 192.168.0.224 Ready GB10 Blackwell OS: Talos Linux v1.12.6 (immutable, API-managed, no SSH) Runtime: containerd Kubernetes: v1.35.3 API: https://k8s.edgeprime.io:6443 (VIP: 192.168.0.250)
INVENTAIRE DES NŒUDS
Spécifications matérielles de chaque nœud du cluster — des PC de bureau reconvertis à un NVIDIA DGX Spark.
| Nœud | CPU | Cœurs | RAM | Disque | GPU | OS |
|---|---|---|---|---|---|---|
| controlplane-100 (192.168.0.100) | AMD A8 PRO-7600B | 4 | 16 GB | 256 GB SSD | — | Talos v1.12.6 |
| controlplane-101 (192.168.0.101) | Intel Pentium G3250T | 2 | 8 GB | 256 GB SSD | — | Talos v1.12.6 |
| controlplane-102 (192.168.0.102) | Intel i5-4690K | 4 | 16 GB | 512 GB SSD | — | Talos v1.12.6 |
| worker-103 (192.168.0.103) | Intel i5-7500T | 4 | 24 GB | 512 GB NVMe | — | Talos v1.12.6 |
| worker-104 (192.168.0.104) | Intel i5-7500T | 4 | 24 GB | 512 GB NVMe | — | Talos v1.12.6 |
| gpu-105 (192.168.0.151) | AMD Ryzen 5 5600X | 12 | 96 GB | 512 GB NVMe | RTX 2060 + 2× RTX 3060 Ti | Talos v1.12.6 |
| worker-106 (192.168.0.150) | Intel i7-4790K | 8 | 16 GB | 128 GB SSD | — | Talos v1.12.6 |
| gx10 (192.168.0.224) | NVIDIA Grace | 20 | 128 GB | 1 TB | GB10 Blackwell | Ubuntu 24.04 (DGX OS) |
| Total cœurs | 58 | Total RAM: 328 GB | ||||
Composants principaux
Talos Linux
productionMinimal, immutable Linux distribution designed specifically for Kubernetes. No SSH, no shell — managed entirely through API.
Rôle : Node operating system for all Talos cluster nodes (3 control plane + 3 workers, including 1 GPU worker)
containerd
productionIndustry-standard container runtime with low overhead and broad compatibility.
Rôle : Container runtime on all nodes
Kubernetes
productionProduction-grade container orchestration system for automating deployment, scaling, and management.
Rôle : Core orchestration platform running v1.35.2