What I run on my homelab 2

•0 likes

A tour of my current homelab: pfSense, Proxmox cluster, HA stacks, media services, observability, AI experiments, and how this all fits into my house network.

You know that feeling when a “temporary” setup somehow becomes a whole ecosystem?
That’s what happened to my homelab.

Earlier this year I wrote about my first proper homelab setup: a single Proxmox node, a handful of LXCs, a NAS, and a wish to expand. Since then things escalated pretty well. I now have a pfSense firewall/router now, a managed TP-Link switch, two Proxmox nodes, a dedicated backup server, a TrueNAS NAS, and a small set of LXC containers where I run my self-hosted services. You can read more on this article What I run on my homelab

The diagram attached to this article is the current snapshot of the madness I built. Let me walk you through it, house, network, servers, and all the nice little choices that turned into a proper home infrastructure.

From “one box” to “mini data center”

The first version of my homelab was classic:

  • ISP modem in router mode
  • Consumer router doing everything
  • A switch just serving as a fancy cable dongle
  • One Proxmox node running all the things

It worked, but I wanted more.

Some issues, bad network performance, random router behavior, and the classic "why is Jellyfin buffering now?" moments like this pushed my hobby into something greater.

So I made four big moves:

  1. Completely re-configured the ISP modem and put a pfSense machine in front of everything.
  2. Properly configured my TP-Link managed switch.
  3. Properly configured my original proxmox host into what it is today: PVE #1 – the compute powerhouse.
  4. Brought in PVE #2 – a warm node/VM Farm, plus a separate TrueNAS and Proxmox Backup Server (PBS) machines running into some scrap hardware and some china findings.

Now the homelab is not just “a server in a corner.” It’s a small, reasonably structured network that behaves like a cut-down version of what I work with professionally, only noisier and closer to my work place. Thankfully, I'm used to it nowadays. I can't live without fan noise anymore.

So, my current setup looks like this today: My Homelab

The house network: internet hits everything

Let’s start at the top of the diagram.

pfSense as the gatekeeper

The flow goes like this:

ISP Modem (Bridge Mode) → pfSense Router/Firewall → TP-Link TL-SG3428 Managed Switch → everything else

pfSense is where I:

  • Terminate the WAN connection
  • Run firewall rules
  • Do DNS tweaks
  • Configure my VPN tunnels

It replaces the ISP router completely. The modem is just a dumb bridge, which means fewer surprises and more control. Also: no more random firmware “upgrades” and duplicated NAT and COGN.

Wi-Fi: ASUS mesh and a miriad of smart stuff

From pfSense and the switch, I feed an ASUS RT-AX82U as my main Wi-Fi/AP and a second AX82U as a mesh node.

On top of that Wi-Fi layer I have all the Wi-Fi Clients (ASUS Wi-Fi Network). Which usually means:

  • Cameras
  • Smart switches
  • Amazon Alexa devices
  • Phones, tablets, laptops, and other gadgets

Nothing exotic, but the important part is: Wi-Fi is just another “edge network” behind pfSense, not the core of the infrastructure.

I also run a few devices wired thru my APs for better connection:

  • A TV on the main router
  • A TV wired to the mesh node
  • And a PlayStation and Nintendo Switch

Wired stuff: PCs, servers, consoles, and TVs

Hanging off the managed switch I’ve got:

  • My work PC
  • My personal notebook
  • My wife’s gaming PC
  • A smart LAN power strip (because I’ve learned to respect my electricity bill)

And then, of course, the serious stuff:

  • Proxmox PVE #1 – the big green bar in the diagram
  • Proxmox PVE #2 – the warm node
  • NAS Server (TrueNAS)
  • Proxmox Backup Server (PBS)

The switch is the silent hero here. I get proper VLAN support, link speeds I can trust, and clean wiring into the rack instead of a bundle of sadness hanging off the router.

Proxmox PVE #1 – the container farm

PVE #1 is labelled in the diagram as:

“Compute powerhouse, Container farm, Core node”

That’s not an exaggeration. This node runs almost everything that feels “core” to my homelab life. And because my brain likes structure, I grouped the LXCs into three stacks:

  • Media Stack
  • Observability Stack
  • Infrastructure Services

Let’s walk through them.

Media Stack (LXC) – everything I watch and forget to finish

The Media Stack is all the orange boxes:

  • 102 – Jellyfin: the central media server.
  • 103 – qBittorrent: the download workhorse.
  • 104 – Prowlarr, 105 – Radarr, 106 – Sonarr, 107 – Jellyseerr: the automation gang, indexers, movies, TV shows, and request front-end.
  • 108 – Bazarr: subtitles, because switching audio tracks every time gets old.
  • 112 – SMB NAS (Old NAS): a legacy share that refuses to die; still holds old media and backups.
  • 126 – Scraparr: because observability is key.

This entire stack lives in containers, sits on top of the TrueNAS for storage.

If you’re starting a homelab, this is usually where the fun begins. You set up Jellyfin or Plex “just to test,” and suddenly you have a bunch of stuff.

Observability Stack (LXC) – because even homelabs deserve graphs

The yellow block is where my inner SRE wakes up:

  • 113 – Prometheus
  • 116 – Prometheus PVE Exporter
  • 117 – Grafana
  • 111 – MySpeed: for continuous speed tests, damn you ISP.
  • 124 – myIP: my IP watcher and cool tools.

Prometheus scrapes everything: PVE nodes, Kubernetes stuff, some services in HAAT, even the Nobara gaming VM. Grafana then turns it into pretty dashboards that I stare at while pretending this is normal home behavior.

Is it overkill for a house? Absolutely.
Is it satisfying to see the latency spike when someone starts streaming 4K? Very.

Infrastructure Services (LXC) – the glue of the homelab

The green block is my “infra shelf”:

  • 100 – Homepage: a simple internal dashboard with links to everything.
  • 101 – Pi-hole: network-wide ad blocking and DNS control.
  • 109 – PostgreSQL and 110 – Redis: shared database/cache for some of my apps and experiments.
  • 118 – Ubuntu Sandbox: I break things here first.
  • 119 – Gitea and 120 – Gitea Mirror: my self-hosted git and a mirror for external repos.
  • 121 – OpenWebUI: for playing with local models and my AI farm.
  • 122 – Jenkins: old friend for CI jobs and random automations - This is responsible for HAAT deploy.
  • 123 – Nginx Proxy Manager: reverse proxy with a nice UI.
  • 125 – WatchYourLAN: small network scanner that warns me when new stuff appears.
  • 128 – ntfy: push notifications for my phone (backups done, disks sad, etc.).
  • 129 – cloudflared: tunnel agent linking my services to the outside when needed.
  • 130 – Netvisor: my network visualization helper.

This is the “if this node explodes, I actually notice” tier.
Pi-hole and Homepage are probably the two things that make the network feel “mine” every single day.

HAAT – my own nerdy platform, in purple

If you scroll down in the diagram you’ll see a long purple bar:

127 – HAAT (VM – k8s)

HAAT is my High Availability Agent Toolkit project, a personal platform for playing with AI agents, research workflows, and multi-service orchestration. Instead of keeping it as one big monolith, I run it as a small Kubernetes cluster inside a VM. HAAT ended by being way more than just an agent power house, it is where I automate and improve my life using my SWE skills.

Inside that HAAT world, I have:

  • Postgres, Redis, Redis Exporter
  • Backend and Frontend
  • Cortex Streams: This is cool, it r eceives events, interprets with AI and take actions if required.
  • Celery Worker and Celery Beat: For async thingies.
  • MinIO for object storage
  • Tempo and Tempo Bucket Init for tracing
  • OTEL Collector
  • HAAT Sandbox: My safe space for my custom coding agents.
  • HAAT Tor Proxy: For when I need privacy.
  • Prometheus (HAAT)
  • A couple of helper services like Portainer and Nginx

The idea is simple: treat HAAT like a real product:

  • It has its own observability stack.
  • It has both ephemeral compute and persistent state.

Which leads to the little sibling on the right side of the diagram:

127 #2 – HAAT (persistent node)

That node keeps the durable pieces: Postgres, Redis, and others somewhere slightly safer. It’s not real multi-region HA, of course, but it adds separation between “things that can be blown away” and “things I really don’t want to lose.”

I also run Celery Worker and Celery Beat on HA, since protocols can never die, they run everytime, and needs to be safe.

This piece runs on my Proxmox PVE #2, which you will read below.

Proxmox PVE #2 – the warm HA, gaming box and AI farm

Scroll further down and you’ll spot Proxmox PVE #2:

“High Availability warm node, Non-critical VM farm”

This node has two personalities.

Part 1 – Nobara OS VM

First, it runs a Nobara OS VM, which is my:

  • Development tooling box
  • Gaming Station (Proton) for Steam and friends
  • Node Exporter for metrics
  • AI Farm (vLLM) for local model experimentation

Yes, the gaming VM is monitored like a server.
Yes, I have Grafana panels showing GPU usage while I play.
No, I’m not saying that’s healthy behavior, but it’s fun.

This entire VM is powered by a GPU RTX 5070TI passthru.

Part 2 – Warm HA zone

Second, PVE #2 acts as a warm HA zone:

“<Any node requiring HA>”

Basically, any VM that I want somewhat redundant or easily movable can live here. If I need to offload a service from PVE #1 for maintenance or experiments, PVE #2 is ready to host it. Or even if PVE #1 blows up, PVE #2 can take the lead.

For HAAT, this node also runs that persistent stack I mentioned:

  • Postgres
  • Redis
  • Redis Exporter
  • Cortex Streams
  • Celery Worker
  • Celery Beat

So the picture looks like this:

  • PVE #1: main container farm, media, infra, and one HAAT k8s node
  • PVE #2: warm node, gaming/AI VM, and HA HAAT services

Is it true high availability? Not really.
Is it “my house is down but at least I know why” availability? Very much.

A serious note: This setup has flaws, since I run a cluster of only 2 nodes, sometimes I can lose quorum. I intend to add a qdevice in the near future.

Storage and backups

On the right side of the diagram there are two blue/green boxes:

  • NAS Server (TrueNAS)
  • Proxmox Backup Server (PBS)

TrueNAS – the serious storage

TrueNAS holds:

  • Media libraries
  • Important documents
  • VM images and templates
  • Random archives

It speaks SMB to the old NAS share container, NFS to some of the hosts, and behaves like the grown-up of the group.

PBS – the “please don’t make me restore from scratch” server

PBS does scheduled backups of:

  • Proxmox VMs and LXCs
  • Some critical datasets via client-side backup
  • Important configurations

When I first set this up, I underestimated how comforting it is to get a notification from PBS saying “backup finished” instead of silently hoping nothing breaks.

Homelabs are great for learning backup strategies the hard way. I’d rather not repeat that part.

Daily life on this homelab

So what does this all give me, in practice?

  • The media stack keeps movies and shows flowing to every screen in the house.
  • Pi-hole and WatchYourLAN keep the network clean and help me see when devices misbehave.
  • Grafana dashboards show CPU, RAM, disk, GPU, and even network throughput—yes, I look at them way more than necessary.
  • HAAT runs research flows, AI experiments, important automations and silly ideas that sometimes turn into serious tools.
  • Nobara VM lets me jump from coding to gaming to AI benchmarks, AI coding, all without switching machines.

It also taught me a lot about:

  • Network design at home scale
  • Power and UPS planning (shout-out to the Intelbras no-breaks)
  • How quickly “I’ll just add one container” turns into a full blown service catalog

Tips if you’re thinking “I want this, but smaller”

You don’t need two Proxmox nodes and a firewall box to start. Honestly, most people shouldn’t start that way.

If you’re at the “curious” stage, here’s a softer path:

  1. Start with one machine
    Old desktop, mini-PC, whatever. Install Proxmox or even just plain Linux with Docker.

  2. Pick two or three services that make your life better
    My short list would be Pi-hole, Jellyfin, and some sort of backup.

  3. Monitor something, even if it’s small
    A simple Prometheus + Grafana pair can teach you a lot about resource usage and “why is this slow?”

  4. Add structure slowly
    Group services logically like I did with Media, Observability, and Infra. It makes growth less chaotic.

  5. Treat your homelab as a playground, not just a mini-production
    Break things, rebuild, try other stacks. The whole point is to learn and have fun, not to recreate a Tier-4 data center next to your sofa.

  6. Play with network
    Create and separate VLANS, understand important concepts, play with firewalls. You decide, but do.

Where this homelab goes next

The current setup feels surprisingly stable:

  • pfSense keeps the front door under control.
  • PVE #1 runs the daily-use services.
  • PVE #2 handles gaming, AI, and “if this crashes it’s fine” experiments.
  • TrueNAS and PBS keep my future self from swearing too much.

Next steps will probably be:

  • Add a qdevice for my proxmox cluster, or simply a new big node.
  • Playing more with GPU workloads on the AI Farm
  • Making HAAT smarter and more self-observing
    • I got a huge backlog for it.
  • Maybe adding more automation around on/off schedules to save power
  • Upgrading my network stack. I'm really looking forward for some ubiquiti devices.

For now, though, this is my homelab. It’s chaotic in my own way, overbuilt for a house, and exactly the kind of environment where I can test ideas without begging a cloud bill for mercy.

If you’re a software engineer, a homelaber, or just someone who enjoys cables a little too much, I hope this gave you a few ideas—and maybe the tiny push to sketch your own version of that diagram and start building.