OPERATIONAL · 34 services · 10 nodes · 10 guests

Home Lab NJ

A resilient, enterprise-grade sandbox for self-hosting, cloud architecture, and data orchestration — three Proxmox nodes, a seven-node Kubernetes cluster, and ~109 TB across two storage arrays, all running from a rack in New Jersey.

Explore architecture Live status

0vCPU · Proxmox

0 GBRAM · fleet

0 TBstorage · 2 arrays

0live services

0Proxmox guests

Infrastructure & Hardware

Three pillars, run like production.

Nothing here is a toy. The compute is highly available, every guest is backed up nightly, and the network is segmented into isolated VLANs — the same patterns you'd ship at work, scaled to a home rack.

Compute Cluster

HA Kubernetes on Proxmox

A 3-node Proxmox VE cluster hosts a 7-node Talos Kubernetes cluster — a control-plane and a worker on every host, plus a dedicated GPU worker — so any single host can fail without losing the cluster. All workloads are declarative and reconciled by ArgoCD.

HypervisorProxmox VE 9.2 ×3

OrchestrationTalos · k8s v1.36

Topology3 control-plane · 4 worker

AcceleratorIntel Arc (Battlemage) passthrough

GitOpsArgoCD · app-of-apps

Storage Pool

~109 TB across two arrays

Two UniFi UNAS arrays. The primary serves dedicated shares over NFS to the cluster and SMB to the LAN, with no-root-squash isolation for Kubernetes; the second holds the long-term backup archive.

Capacity~44 TB + ~65 TB

Sharesdata · cloud · media

BackupsProxmox Backup Server · TimeMachine

ProtocolNFSv3 iso + SMB

Guest backupsnightly · deduplicated

Networking

UniFi SDN & VLANs

A UniFi-driven software-defined network routes and switches everything, with traffic segmented into isolated VLANs. Public services reach the internet through a single Cloudflare tunnel — no ports forwarded.

FabricUniFi SDN

Segmentsmgmt · storage · k8s · guest

IngressMetalLB · ingress-nginx

EdgeCloudflare Tunnel

TLScert-manager · Let's Encrypt

Proxmox Environment

Three hosts, ten guests.

Proxmox VE 9.2 clusters three small-form-factor hosts. Nearly everything runs as Talos VMs; the only other guests are the backup server, the datacenter manager, and an out-of-cluster admin workstation that stays up even when Kubernetes doesn't.

pve-node-01

Ryzen 9 9955HX

CPU16C / 32T

Memory64 GB

Taloscontrol-plane · worker

Alsoadmin CT (HA) · PDM

pve-node-02

Ryzen 9 9955HX

CPU16C / 32T

Memory64 GB

Taloscontrol-plane · worker

AlsoProxmox Backup Server

pve-node-03

Core Ultra 9 285HX

CPU24C / 24T

Memory96 GB

Taloscontrol-plane · worker

AlsoGPU worker · Arc passthrough

ID	Guest	Type	Host	vCPU	RAM	Role
8001–8003	talos-cp	VM	node-01/02/03	2	8 GB	Kubernetes control plane (etcd quorum)
8011–8013	talos-worker	VM	node-01/02/03	4	16 GB	General workloads
8014	talos-gpu	VM	node-03	8	32 GB	GPU worker — Plex transcoding, local LLMs
103	proxmox-backup-server	VM	node-02	4	16 GB	Nightly backups to two NAS datastores
100	claude-code	LXC	node-01	4	32 GB	Admin workstation, kept outside the cluster; HA-managed
102	datacenter-manager	LXC	node-01	2	2 GB	Proxmox Datacenter Manager

Guests are defined in OpenTofu; Talos machine configs and every cluster app live in Git. Hosts are configured with Ansible, and a network UPS shuts the cluster down cleanly through NUT.

Hosted Applications

Thirty-four services, all self-hosted.

Almost everything below runs on the Kubernetes cluster, deployed from Git; a few pieces run directly on Proxmox or the storage array. Filter by what you're curious about.

Environment Snapshot

Where the rack stands.

Storage use, backup coverage, and time since the last full outage — measured September 2026 and refreshed as the environment changes.

Storage pool utilization

2 × UNAS · ~109 TB usable

—

used

—days since the last full outage

last: 2026-09-09, a false-positive UPS shutdown · read the report

Nightly · 9 guestsbackup schedule

vzdump → Proxmox Backup Server · every guest except PBS itself