Omni

For this cluster I am taking advantage of omni from Sidero Labs to manage my Talos installation. The home-ops clusters I've reviewed rely heavily on talosctl to manage their clusters but I wanted to skip a few steps and for $10/month I figured it was worth a shot. After getting more familiar with Talos and Omni I do think you can live with out it but I am enjoying the auth and VPN access that comes out of the box.

The Journey Begins

At this point I've gotten started but documenting as I go endes up being a mess so here's the goods:

I'll recap how the story went so far but first some TODOs:

• Bootstraping flux sets those limits but I'm not sure why

Try:

WARNING Wiping disks is for ceph. They wipe /dev/nvme# which happens to change all the time when talos re-images the OS. Make sure these paths are correct before running this!!!

TODO Devcontainer needs helm plugin install https://github.com/databus23/helm-diff because helm plugin list is missing diff and if the initial apps fail they complain the second time that you need this diff thing!

task omni:validate                # schema-check the cluster template first
task omni:sync                    # provision/assign the nodes
task flux:install-helm-apps       # cilium + coredns + csr-approver + spegel (pre-Flux core)
task rook:wipe-disks-talosm01     # each of these prompts before it erases anything
task rook:wipe-disks-talosm02
task rook:wipe-disks-talosm03
task flux:bootstrap               # needs age.key — run from the workstation

The helm-apps step used to be task talos:install-helm-apps. The whole talos: namespace was talhelper-based and was removed in the 2026-09-22 taskfile audit — Omni owns the machine config, so a raw talosctl upgrade would fight it. Upgrades go through task omni:sync; see .agents/runbooks/talos-version-upgrade.md.

TODO See Rook task file and add RookDiskWipe in which was needed to get bluestore partition off of OSDs