scyto

Building a TrueNAS sysext with Claude Code — an annotated session

This is a lightly-edited, sanitized transcript of a real Claude Code session: building a systemd-sysext that adds MemryX MX3 AI-accelerator support to TrueNAS SCALE, from an empty repo all the way through hardware bring-up with Frigate.

It's preserved to show the workflow — the prompts, the back-and-forth, and how Claude scaffolded the repo, drove its own CI/CD, and debugged a stubborn hardware

scyto

pve-nas1 — PCIe Device Inventory & Passthrough Configuration

Reference snapshot of the host's PCIe layout after the May 2026 boot-pool migration and topology shuffle. Use this as the source of truth when rebuilding VM passthrough configs or updating vfio-pci bindings.

---

Device Inventory

Bus order, identifying who runs each device on the host and how that binding is enforced.

scyto

v0.1 2025-12-23

this has not been validated end-to-end by anyone including me, so be warned :-)

Latest

I moved from squeezelite to multiroom-audio container (i like this approach better)

To Do

see if i can replicate same sound card setup with mix of pulseaudio and pipewire (alsa and pipewire was a bust)

CasaTunes Conversion to Music Assistant

scyto

Enable any LAN client to access mesh

Version 0.5 (2025.04.29)

I have other devices that need to access the ceph mesh that are on my LAN. This gist is only needed if you want LAN clients to access the ceph mesh.

Goals

• let any client on LAN access the mesh
• avoid setting static routes on my router
• enable support for routing topology changes without having to reconfigure router

scyto

Thunderbolt Mesh Cluster Setup Guide (Proxmox + FRR + BGP)

---

📁 Table of Contents

• 1. Network Overview
• 2. Interface Overview (All Nodes)
• 3. Stage 1 — Internal Mesh VM Routing Only
  • 3.1 Configure Network Interfaces
• 3.2 Install FRRouting

scyto

Thanks for hanging tight —
Here’s the corrected, production-quality version based on everything you asked:

---

🛠 Thunderbolt Mesh Setup – Staged Guide (Proxmox + FRR + BGP)

---

📦 Stage 1 — Internal Mesh VM Routing Only

scyto

Give VMs Accesss to Ceph Mesh (routed not bridged access)

Version 0.9 (2025.04.29)

Routed is needed, you can't jut simply bridge en05 and en06 and have VMs work, bridging seems to not work on thundebolt interfaces, at least i could never get the interfaces working when bridged and it broke the ceph mesh completely.

tl;dr can't bridge thunderbolt interfaces

Goal

scyto

CephFS Mounting for Docker VMs (first draft)

2025.04.27 - currently untested e2e this was made from my raw notes by chatgpt, so erros and hallucianation may have crept in :-)

This document describes the clean, final method to mount a CephFS filesystem for Docker VMs across your cluster.

Assumptions:

• you have a working cephFS volume called docker (out of scope)
• that you can see this volume just fine mounted on all 3 pve nodes (if you can't then this is never going to work)
• that you are using the IPv6 version of my ceph proxmox setup (not critical, just sawp out IPv6 for IPv4 address below)

scyto

📚 Proxmox FRR OpenFabric IPv6 Initial Setup (fc00::/128 Design)

---

🔢 Overview

This document describes the original setup to establish an FRR (Free Range Routing) OpenFabric IS-IS based IPv6 routed mesh over Thunderbolt networking between Proxmox nodes, using static /128 loopback addresses in the fc00::/8 ULA space.

This provided:

scyto

IF YOU FIND THIS GIST THIS IS A ROUGH GUIDE TO CHANING CEPH IP AND SUBNETS - MY ADVICE IS DONT IT IS BAD EVERR TIME AND THIS IS ROUGH STEPS IN REALITY ITS MISSING SOME STEPS

🌟 Proxmox Ceph IPv6 Monitor IP Migration Best Practices

I learn't an important lesson today, never ever remove ms_bind_ipv4 = false from ceph.conf or the cephFS will be fucked. note also recreating the mgrs and mds seems advisable too

only ever reboot one node, if that doesn work or you see libceph error storm when it reboot - solve that first (make sure no wrong mons defined in storage.cfg or ceph.conf)