SimpleK8s Controller

Description

The SimpleK8s Controller automates distro updates and reboots across a whole Kubernetes cluster. It runs as a privileged DaemonSet (one pod per node) with two jobs:

Distro updates: each pod checks the signed release repo on a schedule, verifies every artifact against the GPG-signed sha256 index, and stages new kernels on its own boot partition. Nodes only ever consider their own board flavor (x86-64, rpi4, rpi5).
Node reboots: a small HTTP API schedules reboots with a global queue — workers before control planes, at most one control plane at a time — with PodDisruptionBudget-aware, concurrency-limited drains and one Kubernetes Event per state transition.

Reboots only ever start inside a maintenance window you configure; by default nothing reboots automatically.

Installation

$ helm install simplek8s-controller oci://ghcr.io/simplek8s/charts/simplek8s-controller \
    --namespace simplek8s --create-namespace

The code and the design document live at github.com/simplek8s/simplek8s-controller.

Configuration

One flat ConfigMap (simplek8s/simplek8s-controller):

Key	Default	Meaning
`updates.url`	`https://dl.simplek8s.org/simplek8s/stable`	release channel (`dev`/`rolling`/`stable`) or custom URL
`updates.preserve`	`3`	staged kernels to keep per node
`updates.max-percent-usage`	`75`	boot-partition usage cap
`updates.windows`	`["@every 12h"]`	when update work (checks, staging) may run
`reboots.windows`	`[]`	when reboots may start (empty = opt-in; nothing auto-reboots until you set it)
`reboots.max-concurrent`	`1`	parallel reboots
`reboots.drain-timeout`	`10m`	stuck drains fail instead of hanging

Node annotations

Per-node state lives in annotations, so it survives restarts:

simplek8s.org/next-kernel: the node’s boot goal (a release ts). Point it at a staged version and the local pod re-points the bootloader; while reboots.windows is open, a node whose goal differs from the running kernel becomes reboot-eligible by derivation and is enqueued automatically.
Reboot state (requested → draining → rebooting → completed|failed) tracks an in-flight reboot.

Reboot API

There is no Service by design — reach it with a port-forward:

$ kubectl port-forward -n simplek8s daemonset/simplek8s-controller 1880:8080

$ # Schedule a reboot:
$ curl -X POST localhost:1880/api/v1/reboots/my-node
$ # Inspect it:
$ curl localhost:1880/api/v1/reboots/my-node
$ # Cancel it:
$ curl -X DELETE localhost:1880/api/v1/reboots/my-node

Manual updates and rollback

For one node without (or outside) the Controller, use nodectl — it writes the same files, and the Controller adopts a CLI-staged default through the normal next-kernel comparison once annotated.

Rollback is one annotation edit: set simplek8s.org/next-kernel back to the running (or any preserved) version and reboot — old kernels stay on the boot partition under retention.