Skip to content

Roadmap#

concept

Seven phases, each with a test that proves it is done. Every phase builds in a virtual machine first; GPU and fabric validation move to real hardware once the layers exist.

Key decisions#

Decision Choice Why
Base Own base, not a fork Forking Talos is fastest, but the result is Talos renamed. k8or-os studies Talos (machined, its config API, its extension model) as a reference.
Build system Buildroot now, Yocto later Buildroot builds toolchain, kernel and minimal userland from source in under an hour. Yocto takes over with several hardware targets, long-term maintenance and SBOMs.
C library glibc over musl NVIDIA's user-space stack is glibc-linked; glibc from day one avoids a second C library on GPU nodes.
Init BusyBox first, then k8or-init BusyBox gets early phases booting; k8or-init then owns layer verification, mounting, machine config and service supervision.
Images erofs + dm-verity Compressed, read-only, fast to mount, verified on every read, for the base and every layer.
Kernel 6.12 LTS, tightly configured Only what the node needs, with GPU modules built against it in the same pipeline.

Phases#

Toolchain & kernelfoundation

glibc cross-toolchain and the k8or 6.12 LTS kernel configuration, built with Buildroot.

boots in QEMU to a minimal initramfs shell
Minimal userlandbase

PID 1, device handling, networking, time sync, read-only erofs root, A/B partition layout.

boots a read-only root with networking and time sync; A/B slots present
Container layerkubernetes

containerd, runc, CNI plugins and kubelet, with cgroup v2 throughout.

a VM joins a kubeadm cluster and runs a pod
API-driven configno shell

The k8or machine-config API replaces shell and SSH access.

a node is configured and upgraded with no shell or SSH
Fabric & GPU layersacceleration

rdma-core and RoCE v2 tuning; NVIDIA open kernel modules, Fabric Manager, container toolkit and DCGM, as signed layers.

the GPU layer loads on its matching kernel and is refused on a mismatch; nvidia-smi and ibv_devinfo pass inside a pod
Trust & updatessecurity

Unified kernel image, Secure Boot, dm-verity, TPM-measured boot, signed A/B updates.

a signed update applies, and a failed boot rolls back automatically
Hardware validationH100 · RoCE v2

Real GPU servers, real fabric, and the AI-Ops layer under real faults.

nccl-tests all-reduce reaches the expected bus bandwidth, the GPUDirect RDMA path is verified, and Underlay explains injected faults correctly

Beyond phase 6

The k8or base container image, further accelerator layers (AMD ROCm, Intel Gaudi), and a move to Yocto for long-term maintenance.