Kairos v4.2.0: In-Memory Boot, MaaS Support, and Hadron Extension Layers
Kairos v4.2.0 is now available.
This release opens up new places to run Kairos and new ways to extend it: boot entirely from RAM while still keeping persistent config and data on disk, deploy through Canonical MaaS, and extend Hadron images with the first extension layers. It also brings initial NVIDIA DGX Spark support, a hardening baseline for Hadron, and steady progress on confidential computing.
Boot entirely from RAMโ
Kairos can now boot with its root filesystem copied entirely into RAM, so nothing on the boot disk is touched at runtime โ closer to how ESXi boots than to a typical live CD, but with persistent state still available if you want it.
It's enabled with the kairos.ram boot parameter. On its own, that's a fully diskless boot. Add-on flags let you keep real COS_OEM and COS_PERSISTENT partitions on disk, so configuration and user data survive a reboot even though the OS itself runs from memory:
kairos.ram.create_partitions[=/dev/sdX]โ provision OEM/persistent partitions on first bootkairos.ram.wipeโ explicit consent to touch a disk that has foreign partitions on itkairos.ram.oem=<MiB>/kairos.ram.persistent=<MiB>โ size the partitions
This is aimed at appliances and edge nodes where you want zero disk wear and a boot that's effectively a fresh, known-good state every time, without giving up persistent configuration. It shipped across kairos-sdk (in-RAM detection) and immucore (the boot workflow itself, plus a clearer failure screen if something goes wrong), and it's documented in the netboot guide.
The first Hadron extension layersโ
Hadron images can now be extended with extension layers: official packages and host-bound components โ think firmware, kernel modules, things that don't belong as generic OS packages โ delivered either as sysexts you can add or remove at runtime without rebuilding the image, or as build-time layers baked in when the image is built.
The layers themselves live in hadron-layers (currently drbd, fwupd, git, and gpg), with a matching hadron-firmware repo for firmware. They're published as OCI images and listed on hadron-linux.io/components.
This is the first version of the system, and we're treating it as experimental: try it, and tell us what layers you need. Community-contributed layers are the direction we want to take this next.
MaaS supportโ
Kairos now supports deployment through Canonical MaaS (Metal as a Service), with a dedicated MaaS data source for pulling configuration during provisioning. If you already manage bare metal through MaaS, you can bring Kairos into that same workflow instead of running a separate provisioning path for it. See the MaaS installation guide for setup.
NVIDIA DGX Spark supportโ
Kairos adds initial support for NVIDIA DGX Spark boards (GB10 Grace-Blackwell), including bundled signing keys, board-specific console options, and package mappings in kairos-init.
This lands alongside a broader cleanup of how Kairos builds NVIDIA images: AuroraBoot now emits individual partition images for flashing directly, replacing per-board custom scripts with one standard flow โ the same work that already made building for boards like the AGX Orin more uniform. See the DGX Spark build guide for the steps.
A hardening baseline for Hadronโ
Hadron images now ship a real STIG-inspired hardening baseline: an sshd configuration drop-in, around 30 sysctl settings aligned with GPOS guidance, SYN cookies enabled by default, a blocklist for legacy network-protocol kernel modules, and tightened login.defs (password aging, UMASK, login logging).
This is a baseline, not full STIG compliance โ PAM hardening (password complexity, history, pam_wheel) and a few other findings didn't make it into this release, and are left for derivative images and your own cloud-config to add on top.
Confidential computing: real progress, not there yet on Hadronโ
Confidential computing support without Hadron is done and documented in this release. Getting it working on Hadron itself is a separate, harder problem: our Hadron-built kernel doesn't currently boot under AMD SEV-SNP, and we haven't been able to pin down why on cloud VMs alone. We're actively trying to get our hands on dedicated SEV-SNP-capable bare metal to debug it properly. If you have access to hardware like this and want to help, reach out โ this is exactly the kind of thing that needs a real machine, not another cloud VM.
Cloud-config changesโ
Two smaller but useful additions to what you can do in a Kairos cloud-config:
- Templating for
config_urlโ the URL Kairos fetches its config from at boot can now use templating, useful when the same image needs to resolve a different config per machine. gidfor users โ you can now specify a group ID alongside a user, not just the username and UID.
Kairos v3 reaches end of lifeโ
With this release, we've also retired the last Kairos v3 documentation still published on the site (v3.7.2) โ see kairos-docs#668. We keep the latest patch of the three most recent major.minor lines, and v4.2.0 pushes v3 out of that window. If you're still running a v3.x deployment, now is a good time to plan your upgrade to v4.
And moreโ
This release also includes a long tail of fixes worth knowing about if you're upgrading: manual-partition sizing now checks against the disk's actual size, control planes are sorted first during node upgrades even without a node selector, AuroraBoot's reported image size is corrected, a user-UID regression is fixed, duplicate extra-partition handling between ISO and user-data sources is resolved, and unformatted extra-partitions no longer fail with an "unsupported filesystem type" error. Reading directly from /etc/elemental/config.yaml is now deprecated.
Looking aheadโ
Kairos v4.2.0 is about giving you more ways to run Kairos and more ways to shape it: boot from RAM when you want a disk-light appliance, reach for MaaS or DGX Spark when that's where your hardware lives, and use the first Hadron extension layers when the base image doesn't have quite what you need. The Hadron hardening baseline and continued confidential-computing work point at the same underlying goal โ Kairos as a foundation you can trust on hardware you don't fully control.
You can find the full release notes here: Kairos v4.2.0 on GitHub.
Thanks to everyone who contributed to this release.
