You should be familiar with Fedora CoreOS, as this is an OCI image of CoreOS with "batteries included". More specifically, it's an opinionated, custom CoreOS image, built daily with some commonly used tools added in. The idea is to make a lightweight server image including most used services or the building blocks to host them.
WARNING: This image has not been heavily tested, though the underlying components have. Please take a look at the included modifications and help test if this project interests you.
NOTE: formerly named fedora-coreos-zfs
, that version of the image did not offer the nvidia option. Please update with rpm-ostree rebase
.
A generic Fedora CoreOS image image with choice of add-on kernel modules:
- nvidia versions add:
- nvidia driver - latest driver (currently version 535) built from negativo17's akmod package
- nvidia-container-toolkit - latest toolkit which supports both root and rootless podman containers and CDI
- nvidia container selinux policy - allows using
--security-opt label=type:nvidia_container_t
for some jobs (some will still need--security-opt label=disable
as suggested by nvidia)
- ZFS versions add:
- ZFS driver - latest driver (currently pinned to 2.1.x series)
NOTE: currently, zincati fails to start on systems with OCI based deployments (like uCore). Upstream efforts are active to correct this.
Suitable for running containerized workloads on either baremetal or virtual machines, this image tries to stay lightweight but functional for multiple use cases, including that of a storage server (NAS).
- Starts with a Fedora CoreOS image
- Adds the following:
- cockpit
- distrobox
- duperemove
- guest VM agents (
qemu-guest-agent
andopen-vm-tools
) - intel wifi firmware - CoreOS omits this despite including atheros wifi firmware... hardware enablement FTW
- mergerfs
- moby-engine(docker), docker-compose and podman-compose
- snapraid
- tailscale and wireguard-tools
- tmux
- udev rules enabling full functionality on some Realtek 2.5Gbit USB Ethernet devices
- Optional nvidia versions add:
- nvidia driver - latest driver (currently version 535) built from negativo17's akmod package
- nvidia-container-toolkit - latest toolkit which supports both root and rootless podman containers and CDI
- nvidia container selinux policy - allows using
--security-opt label=type:nvidia_container_t
for some jobs (some will still need--security-opt label=disable
as suggested by nvidia)
- Optional ZFS versions add:
- sanoid/syncoid dependencies - see below for details
- ZFS driver - latest driver (currently pinned to 2.1.x series)
- Enables staging of automatic system updates via rpm-ostreed
- Enables password based SSH auth (required for locally running cockpit web interface)
- Disables Zincati auto upgrade/reboot service
Note: per cockpit instructions the cockpit-ws RPM is not installed, rather it is provided as a pre-defined systemd service which runs a podman container.
Hyper-Coverged Infrastructure(HCI) refers to storage and virtualization in one place... So this image primarily adds the virtualization stack.
- Starts with
ucore
to give you everything above, plus: - Adds the following:
- cockpit-machines: Cockpit GUI for managing virtual machines
- libvirt-client:
virsh
command-line utility for managing virtual machines - libvirt-daemon-kvm: libvirt KVM hypervisor management
- virt-install: command-line utility for installing virtual machines
Note: Fedora now uses DefaultTimeoutStop=45s
for systemd services which could cause libvirtd
to quit before shutting down slow VMs. Consider adding TimeoutStopSec=120s
as an override for libvirtd.service
if needed.
These images are immutable and you probably shouldn't install packages like in a mutable "normal" distribution.
Fedora CoreOS expects the user to run services using podman. moby-engine
, the free Docker implementation, is installed for those who desire docker instead of podman.
To maintain this image's suitability as a minimal container host, most add-on services are not auto-enabled.
To activate pre-installed services (cockpit
, docker
, tailscaled
, etc):
sudo systemctl enable --now SERVICENAME.service
Note: libvirtd
is enabled by default, but only starts when triggerd by it's socket (eg, using virsh
or other clients).
NOTE: CoreOS cautions against running podman and docker containers at the same time. Thus, docker.socket
is disabled by default to prevent accidental activation of the docker daemon, given podman is the default.
Podman and firewalld can sometimes conflict such that a firewall-cmd --reload
removes firewall rules generated by podman.
A service is included to mitigate this by monitoring for firewall reload events on dbus and then reloading podman networks. If needed, enable like so: systemctl enable --now podman-firewalld-reload.service
Users may use distrobox to run images of mutable distributions where applications can be installed with traditional package managers. This may be useful for installing interactive utilities such has htop
, nmap
, etc. As stated above, however, services should run as containers.
It's a good idea to become familar with the Fedora CoreOS Documentation as well as the CoreOS rpm-ostree docs. Note especially, this image is only possible due to ostree native containers.
sanoid/syncoid is a great tool for manual and automated snapshot/transfer of ZFS datasets. However, there is not a current stable RPM, rather they provide instructions on installing via git.
ucore
has pre-install all the (lightweight) required dependencies (perl-Config-IniFiles perl-Data-Dumper perl-Capture-Tiny perl-Getopt-Long lzop mbuffer mhash pv), such that a user wishing to use sanoid/syncoid only need install the "sbin" files and create configuration/systemd units for it.
If you installed an image with -nvidia
in the tag, the nvidia kernel module, basic CUDA libraries, and the nvidia-container-toolkit are all are pre-installed.
Note, this does NOT add desktop graphics services to your images, but it DOES enable your compatible nvidia GPU to be used for nvdec, nvenc, CUDA, etc. Since this is CoreOS and it's primarily intended for container workloads the nvidia container toolkit should be well understood.
Note the included driver is the latest nvidia driver as bundled by negativo17. This package was chosen over rpmfusion's due to it's granular packages which allow us to install just the minimal nvidia-driver-cuda
packages.
If you need an older (or different) driver, consider looking at the container-toolkit-fcos driver. It provides pre-bundled container images with nvidia drivers for FCOS, allowing auto-build/loading of the nvidia driver IN podman, at boot, via a systemd service.
If going this path, you likely won't want to use the ucore
-nvidia
image, but would use the suggested systemd service. The nvidia container toolkit will still be required but can by layered easily.
If you installed an image with -zfs
in the tag (or fedora-coreos-zfs
), the ZFS kernel module and tools are pre-installed, but like other services, ZFS is not pre-configured to load on default.
Load it with the command modprobe zfs
and use zfs
and zpool
commands as desired.
Per the OpenZFS Fedora documentation:
By default ZFS kernel modules are loaded upon detecting a pool. To always load the modules at boot:
echo zfs > /etc/modules-load.d/zfs.conf
The default mountpoint for any newly created zpool tank
is /tank
. This is a problem in CoreOS as the root filesystem (/
) is immutable, which means a directory cannot be created as a mountpoint for the zpool. An example of the problem looks like this:
# zpool create tank /dev/sdb
cannot mount '/tank': failed to create mountpoint: Operation not permitted
To avoid this problem, always create new zpools with a specified mountpoint:
# zpool create -m /var/tank tank /dev/sdb
If you do forget to specify the mountpoint, or you need to change the mountpoint on an existing zpool:
# zfs set mountpoint=/var/tank tank
This image is not currently available for direct install. The user must follow the CoreOS installation guide. There are varying methods of installation for bare metal, cloud providers, and virtualization platforms.
All CoreOS installation methods require the user to produce an Ignition file. This Ignition file should, at mimimum, set a password and SSH key for the default user (default username is core
).
You can rebase any Fedora CoreOS x86_64 installation to uCore. Installing CoreOS itself can be done through a number of provisioning methods.
To rebase an Fedora CoreOS machine to the latest uCore (stable):
- Execute the
rpm-ostree rebase
command (below) with desiredIMAGE
andTAG
. - Reboot, as instructed.
- After rebooting, you should pin the working deployment which allows you to rollback if required.
sudo rpm-ostree rebase ostree-unverified-registry:ghcr.io/ublue-os/IMAGE:TAG
IMAGE | TAG |
---|---|
fedora-coreos - stable |
stable-nvidia , stable-zfs ,stable-nvidia-zfs |
fedora-coreos - testing |
testing-nvidia , testing-zfs , testing-nvidia-zfs |
ucore - stable |
stable , stable-nvidia , stable-zfs ,stable-nvidia-zfs |
ucore - testing |
testing , testing-nvidia , testing-zfs , testing-nvidia-zfs |
ucore-hci - stable |
stable , stable-nvidia , stable-zfs ,stable-nvidia-zfs |
ucore-hci - testing |
testing , testing-nvidia , testing-zfs , testing-nvidia-zfs |
Your path to a running uCore can be shortend by using examples/ucore-autorebase.butane as the starting point for your CoreOS ignition file.
- As usual, you'll need to follow the docs to setup a password. Substitute your password hash for
YOUR_GOOD_PASSWORD_HASH_HERE
in theucore-autorebase.butane
file, and add your ssh pub key while you are at it. - Generate an ignition file from your new
ucore-autorebase.butane
using the butane utility. - Now install CoreOS for hypervisor, cloud provider or bare-metal. Your ignition file should work for any platform, auto-rebasing to the
ucore:stable
(or otherIMAGE:TAG
combo), rebooting and leaving your install ready to use.
These images are signed with sigstore's cosign. You can verify the signature by downloading the cosign.pub
key from this repo and running the following command:
cosign verify --key cosign.pub ghcr.io/ublue-os/ucore