Proxmox Homelab Cluster

Three-node Proxmox cluster on Dell OptiPlex 7040s — VM hosting, Traefik + Caddy reverse proxy, and self-hosted services for Heff properties.

Proxmox Homelab Cluster

Projektschwerpunkt

Software

Key Features

Three-node quorum cluster
Two-tier reverse proxy
Self-hosted service catalog

Scope of work

Provide owned infrastructure for side projects (Stagea platform, shop, forums, media) without surprise VPS egress bills:

  • Cluster three used OptiPlex 7040 SFF nodes with Proxmox VE
  • Carve VMs/containers for LubeLogger, Manyfold, Matrix, photo apps, and dev sandboxes
  • Design Traefik (edge) + Caddy (origin) reverse-proxy chain for TLS and routing
  • Document backup, migration, and failure modes for single-node loss
  • Keep RAM/CPU budgets honest for homelab workloads

Out of scope: geo-redundant DR, 24/7 NOC staffing, Ceph storage cluster.

Timeline

PhaseWindowMilestone
Hardware procurementQ4 2025Matched 7040 nodes + SSD layout
Cluster bootstrapQ1 2026Corosync quorum, local-lvm pools
Edge proxyQ1 2026Traefik on dedicated VM — ACME, routing
Origin proxyMar 2026Caddy per service VM — simpler site blocks
Service migrationsMar 2026Stagea, shop, Matrix, media subdomains live

Stack & BOM (hardware)

ItemQtyNotes
Dell OptiPlex 7040 SFF36th-gen i5 class, 16–32 GB RAM each
SSD (boot)3Proxmox OS + local-lvm
SSD/HDD (data)3VM disks — no Ceph budget
Managed switch1Flat lab network; optional NIC separation

Stack & BOM (software)

LayerComponents
HypervisorProxmox VE 8.x cluster
EdgeTraefik — TLS termination, host routing
OriginCaddy — per-VM site configs
WorkloadsDocker stacks, Matrix, static sites
ObservabilityNode-level logs; no full APM

Deliverables

  • Three-node Proxmox cluster with quorum
  • Documented Traefik + Caddy split (why two layers)
  • VMs for primary Heff services
  • Backup procedure for VM disks and configs
  • Automated off-site backup rotation (in progress)

Outcomes

  • Unified hosting for properties previously scattered across ad-hoc VPS trials
  • Predictable failure domain — lose one node, migrate VMs, accept reduced redundancy
  • Reverse-proxy design trades single-file Caddy simplicity for clear edge vs. origin boundaries

Operational constraints

  • RAM is the bottleneck — overcommit until reality arrives
  • Used SFF boxes are quiet but not datacenter-redundant PSUs
  • TLS renewals and DNS must stay aligned with Traefik entrypoints

Build-Log & Kontext

Scope, Zeitplan und Stückliste stehen hier. Für Entscheidungen und längeren Kontext den verlinkten Blogbeitrag lesen.

Blogbeitrag lesen
Completed on: 22. März 2026

© 2026 Heff.world. Alle Rechte vorbehalten.