开源软件名称（OpenSource Name）：

开源软件地址(OpenSource Url)：

开源编程语言(OpenSource Language)：

开源软件介绍(OpenSource Introduction)：

k3OS

k3OS is a Linux distribution designed to remove as much OS maintenance as possible in a Kubernetes cluster. It is specifically designed to only have what is needed to run k3s. Additionally the OS is designed to be managed by kubectl once a cluster is bootstrapped. Nodes only need to join a cluster and then all aspects of the OS can be managed from Kubernetes. Both k3OS and k3s upgrades are handled by the k3OS operator.

1. Quick Start
2. Design
3. Installation
4. Configuration
5. Upgrade/Maintenance
6. Building
7. Configuration Reference

Quick Start

Download the ISO from the latest release and run it in VMware, VirtualBox, KVM, or bhyve. The server will automatically start a single node Kubernetes cluster. Log in with the user rancher and run kubectl. This is a "live install" running from the ISO media and changes will not persist after reboot.

To copy k3OS to local disk, after logging in as rancher run sudo k3os install. Then remove the ISO from the virtual machine and reboot.

Live install (boot from ISO) requires at least 2GB of RAM. Local install requires 1GB RAM.

Design

Core design goals of k3OS are

1. Minimal OS for running Kubernetes by way of k3s
2. Ability to upgrade and configure using kubectl
3. Versatile installation to allow easy creation of OS images.

File System Structure

Critical to the design of k3OS is how that file system is structured. A booted system will look as follows

/etc - ephemeral
/usr - read-only (except /usr/local is writable and persistent)
/k3os - system files
/home - persistent
/var - persistent
/opt - persistent
/usr/local - persistent

/etc

All configuration in the system is intended to be ephemeral. If you change anything in /etc it will revert on next reboot. If you wish to persist changes to the configuration they must be done in the k3OS config.yaml which will be applied on each boot.

/usr

The entire user space is stored in /usr and as read-only. The only way to change /usr is to change versions of k3OS. The directory /usr/local is a symlink to /var/local and therefore writable.

/k3os

The k3OS directory contains the core operating system files references on boot to construct the file system. It contains squashfs images and binaries for k3OS, k3s, and the Linux kernel. On boot the appropriate version for all three will be chosen and configured.

/var, /usr/local, /home, /opt

Persistent changes should be kept in /var, /usr/local, /home, or /opt.

Upstream Distros

Most of the user-space binaries comes from Alpine and are repackaged for k3OS. Currently the kernel source is coming from Ubuntu 20.04 LTS. Some code and a lot of inspiration came from LinuxKit

Installation

Interactive Installation

Interactive installation is done from booting from the ISO. The installation is done by running k3os install. The k3os install sub-command is only available on systems booted live. An installation to disk will not have k3os install. Follow the prompts to install k3OS to disk.

The installation will format an entire disk. If you have a single hard disk attached to the system it will not ask which disk but just pick the first and only one.

Automated Installation

Installation can be automated by using kernel cmdline parameters. There are a lot of creative solutions to booting a machine with cmdline args. You can remaster the k3OS ISO, PXE boot, use qemu/kvm, or automate input with packer. The kernel and initrd are available in the k3OS release artifacts, along with the ISO.

The cmdline value k3os.mode=install or k3os.fallback_mode=install is required to enable automated installations. Below is a reference of all cmdline args used to automate installation

Custom partition layout

By default k3OS expects one partition to exist labeled K3OS_STATE. K3OS_STATE is expected to be an ext4 formatted filesystem with at least 2GB of disk space. The installer will create this partitions and file system automatically, or you can create them manually if you have a need for an advanced file system layout.

Bootstrapped Installation

You can install k3OS to a block device from any modern Linux distribution. Just download and run install.sh. This script will run the same installation as the ISO but is a bit more raw and will not prompt for configuration.

Usage: ./install.sh [--force-efi] [--debug] [--tty TTY] [--poweroff] [--takeover] [--no-format] [--config https://.../config.yaml] DEVICE ISO_URL

Example: ./install.sh /dev/vda https://github.com/rancher/k3os/releases/download/v0.10.0/k3os.iso

DEVICE must be the disk that will be partitioned (/dev/vda). If you are using --no-format it should be the device of the K3OS_STATE partition (/dev/vda2)

The parameters names refer to the same names used in the cmdline, refer to README.md for
more info.

Remastering ISO

To remaster the ISO all you need to do is copy /k3os and /boot from the ISO to a new folder. Then modify /boot/grub/grub.cfg to add whatever kernel cmdline args for auto-installation. To build a new ISO just use the utility grub-mkrescue as follows:

# Ubuntu: apt install grub-efi grub-pc-bin mtools xorriso
# CentOS: dnf install grub2-efi grub2-pc mtools xorriso
# Alpine: apk add grub-bios grub-efi mtools xorriso
mount -o loop k3os.iso /mnt
mkdir -p iso/boot/grub
cp -rf /mnt/k3os iso/
cp /mnt/boot/grub/grub.cfg iso/boot/grub/

# Edit iso/boot/grub/grub.cfg

grub-mkrescue -o k3os-new.iso iso/ -- -volid K3OS

GRUB2 CAVEAT: Some non-Alpine installations of grub2 will create ${ISO}/boot/grub2 instead of ${ISO}/boot/grub which will generally lead to broken installation media. Be mindful of this and modify the above commands (that work with this path) accordingly. Systems that exhibit this behavior typically have grub2-mkrescue on the path instead of grub-mkrescue.

Takeover Installation

A special mode of installation is designed to install to a current running Linux system. This only works on ARM64 and x86_64. Download install.sh and run with the --takeover flag. This will install k3OS to the current root and override the grub.cfg. After you reboot the system k3OS will then delete all files on the root partition that are not k3OS and then shutdown. This mode is particularly handy when creating cloud images. This way you can use an existing base image like Ubuntu and install k3OS over the top, snapshot, and create a new image.

In order for this to work a couple of assumptions are made. First the root (/) is assumed to be an ext4 partition. Also it is assumed that grub2 is installed and looking for the configuration at /boot/grub/grub.cfg. When running --takeover ensure that you also set --no-format and DEVICE must be set to the partition of /. Refer to the AWS packer template to see this mode in action. Below is any example of how to run a takeover installation.

./install.sh --takeover --debug --tty ttyS0 --config /tmp/config.yaml --no-format /dev/vda1 https://github.com/rancher/k3os/releases/download/v0.10.0/k3os.iso

ARM Overlay Installation

If you have a custom ARMv7 or ARM64 device you can easily use an existing bootable ARM image to create a k3OS setup. All you must do is boot the ARM system and then extract k3os-rootfs-arm.tar.gz to the root (stripping one path, look at the example below) and then place your cloud-config at /k3os/system/config.yaml. For example:

curl -sfL https://github.com/rancher/k3os/releases/download/v0.10.0/k3os-rootfs-arm.tar.gz | tar zxvf - --strip-components=1 -C /
cp myconfig.yaml /k3os/system/config.yaml
sync
reboot -f

This method places k3OS on disk and also overwrites /sbin/init. On next reboot your ARM bootloader and kernel should be loaded, but then when user space is to be initialized k3OS should take over. One important consideration at the moment is that k3OS assumes the root device is not read only. This typically means you need to remove ro from the kernel cmdline. This should be fixed in a future release.

Configuration

All configuration is done through a single cloud-init style config file that is either packaged in the image, downloaded though cloud-init or managed by Kubernetes. The configuration file is found at

/k3os/system/config.yaml
/var/lib/rancher/k3os/config.yaml
/var/lib/rancher/k3os/config.d/*

The /k3os/system/config.yaml file is reserved for the system installation and should not be modified on a running system. This file is usually populated by during the image build or installation process and contains important bootstrap information (such as networking or cloud-init data sources).

The /var/lib/rancher/k3os/config.yaml or config.d/* files are intended to be used at runtime. These files can be manipulated manually, through scripting, or managed with the Kubernetes operator.

Sample config.yaml

A full example of the k3OS configuration file is as below.

ssh_authorized_keys:
- ssh-rsa AAAAB3NzaC1yc2EAAAADAQAB...
- github:ibuildthecloud
write_files:
- encoding: ""
  content: |-
    #!/bin/bash
    echo hello, local service start
  owner: root
  path: /etc/local.d/example.start
  permissions: '0755'
hostname: myhost
init_cmd:
- "echo hello, init command"
boot_cmd:
- "echo hello, boot command"
run_cmd:
- "echo hello, run command"

k3os:
  data_sources:
  - aws
  - cdrom
  modules:
  - kvm
  - nvme
  sysctl:
    kernel.printk: "4 4 1 7"
    kernel.kptr_restrict: "1"
  dns_nameservers:
  - 8.8.8.8
  - 1.1.1.1
  ntp_servers:
  - 0.us.pool.ntp.org
  - 1.us.pool.ntp.org
  wifi:
  - name: home
    passphrase: mypassword
  - name: nothome
    passphrase: somethingelse
  password: rancher
  server_url: https://someserver:6443
  token: TOKEN_VALUE
  labels:
    region: us-west-1
    somekey: somevalue
  k3s_args:
  - server
  - "--cluster-init"
  environment:
    http_proxy: http://myserver
    https_proxy: http://myserver
  taints:
  - key1=value1:NoSchedule
  - key1=value1:NoExecute

Refer to the configuration reference for full details of each configuration key.

Kubernetes

Since k3OS is built on k3s all Kubernetes configuration is done by configuring k3s. This is primarily done through environment and k3s_args keys in config.yaml. The write_files key can be used to populate the /var/lib/rancher/k3s/server/manifests folder with apps you'd like to deploy on boot.

Refer to k3s docs for more information on how to configure Kubernetes.

Kernel cmdline

All configuration can be passed as kernel cmdline parameters too. The keys are dot separated. For example k3os.token=TOKEN. If the key is a slice, multiple values are set by repeating the key, for example k3os.dns_nameserver=1.1.1.1 k3os.dns_nameserver=8.8.8.8. You can use the plural or singular form of the name, just ensure you consistently use the same form. For map values the form key[key]=value form is used, for example k3os.sysctl[kernel.printk]="4 4 1 7". If the value has spaces in it ensure that the value is quoted. Boolean keys expect a value of true or false or no value at all means true. For example k3os.install.efi is the same as k3os.install.efi=true.

Phases

Configuration is applied in three distinct phases: initrd, boot, runtime. initrd is run during the initrd phase before the root disk has been mounted. boot is run after the root disk is mounted and the file system is setup, but before any services have started. There is no networking available yet at this point. The final stage runtime is executed after networking has come online. If you are using a configuration from a cloud provider (like AWS userdata) it will only be run in the runtime phase. Below is a table of which config keys are supported in each phase.

Networking

Networking is powered by connman. To configure networking a couple of helper keys are available: k3os.dns_nameserver, k3os.ntp_servers, k3os.wifi. Refer to the reference for a full explanation of those keys. If you wish to configure a HTTP proxy set the http_proxy, and https_proxy fields in k3os.environment. All other networking configuration should be done by configuring connman directly by using the write_files key to create connman service files.

Upgrade and Maintenance

Upgrading and reconfiguring k3OS is all handled through the Kubernetes operator. The operator is still in development. More details to follow. The basic design is that one can set the desired k3s and k3OS versions, plus their configuration and the operator will roll that out to the cluster.

Automatic Upgrades

Integration with rancher/system-upgrade-controller has been implemented as of v0.9.0. To enable a k3OS node to automatically upgrade from the latest GitHub release you will need to make sure it has the label k3os.io/upgrade with value latest (for k3OS versions prior to v0.11.x please use label plan.upgrade.cattle.io/k3os-latest). The upgrade controller will then spawn an upgrade job that will drain most pods, upgrade the k3OS content under /k3os/system, and then reboot. The system should come back up running the latest kernel and k3s version bundled with k3OS and ready to schedule pods.

Pre v0.9.0

If your k3OS installation is running a version prior to the v0.9.0 release or one of its release candidates you can setup the system upgrade controller to upgrade your k3OS by following these steps:

# apply the system-upgrade-controller manifest (once per cluster)
kubectl apply -f https://raw.githubusercontent.com/rancher/k3os/v0.10.0/overlay/share/rancher/k3s/server/manifests/system-upgrade-controller.yaml
# after the system-upgrade-controller pod is Ready, apply the plan manifest (once per cluster)
kubectl apply -f https://raw.githubusercontent.com/rancher/k3os/v0.10.0/overlay/share/rancher/k3s/server/manifests/system-upgrade-plans/k3os-latest.yaml
# apply the `plan.upgrade.cattle.io/k3os-latest` label as described above (for every k3OS node), e.g.
kubectl label nodes -l k3os.io/mode plan.upgrade.cattle.io/k3os-latest=enabled # this should work on any cluster with k3OS installations at v0.7.0 or greater

Manual Upgrades

For single-node or development use cases, where the operator is not being used, you can upgrade the rootfs and kernel with the following commands. If you do not specify K3OS_VERSION, it will default to the latest release.

When using an overlay install such as on Raspberry Pi (see ARM Overlay Installation) the original distro kernel (such as Raspbian) will continue to be used. On these systems the k3os-upgrade-kernel script will exit with a warning and perform no action.

export K3OS_VERSION=v0.10.0
/usr/share/rancher/k3os/scripts/k3os-upgrade-rootfs
/usr/share/rancher/k3os/scripts/k3os-upgrade-kernel

You should always remember to backup your data first, and reboot after upgrading.

Manual Upgrade Scripts Have Been DEPRECATED

These scripts have been deprecated as of v0.9.0 are still on the system at /usr/share/rancher/k3os/scripts.

Building

To build k3OS you just need Docker and then run make. All artifacts will be put in ./dist/artifacts. If you are running on Linux you can run ./scripts/run to run a VM of k3OS in the terminal. To exit the instance type CTRL+a c to get the qemu console and then q for quit.

The source for the kernel is in https://github.com/rancher/k3os-kernel and similarly you just need to have Docker and run make to compile the kernel.

Configuration Reference

Below is a reference of all keys available in the config.yaml

ssh_authorized_keys

A list of SSH authorized keys that should be added to the rancher user. k3OS primarily has one user, rancher. The root account is always disabled, has no password, and is never assigned a ssh key. SSH keys can be obtained from GitHub user accounts by using the format github:${USERNAME}. This is done by downloading the keys from https://github.com/${USERNAME}.keys.

Example

ssh_authorized_keys:
- "ssh-rsa AAAAB3NzaC1yc2EAAAADAQABAAABAQC2TBZGjE+J8ag11dzkFT58J3XPONrDVmalCNrKxsfADfyy0eqdZrG8hcAxAR/5zuj90Gin2uBR4Sw6Cn4VHsPZcFpXyQCjK1QDADj+WcuhpXOIOY3AB0LZBly9NI0ll+8lo3QtEaoyRLtrMBhQ6Mooy2M3MTG4JNwU9o3yInuqZWf9PvtW6KxMl+ygg1xZkljhemGZ9k0wSrjqif+8usNbzVlCOVQmZwZA+BZxbdcLNwkg7zWJSXzDIXyqM6iWPGXQDEbWLq3+HR1qKucTCSxjbqoe0FD5xcW7NHIME5XKX84yH92n6yn+rxSsyUfhJWYqJd+i0fKf5UbN6qLrtd/D"
- "github:ibuildthecloud"

write_files

A list of files to write to disk on boot. These files can be either plain text, gziped, base64 encoded, or base64+gzip encoded.

Example

write_files:
- encoding: b64
  content: CiMgVGhpcyBmaWxlIGNvbnRyb2xzIHRoZSBzdGF0ZSBvZiBTRUxpbnV4...
  owner: root:root
  path: /etc/connman/main.conf
  permissions: '0644'
- content: |
    # My new /etc/sysconfig/samba file

    SMDBOPTIONS="-D"
  path: /etc/sysconfig/samba
- content: !!binary |
    f0VMRgIBAQAAAAAAAAAAAAIAPgABAAAAwARAAAAAAABAAAAAAAAAAJAVAAAAAA
    AEAAHgAdAAYAAAAFAAAAQAAAAAAAAABAAEAAAAAAAEAAQAAAAAAAwAEAAAAAAA
    AAAAAAAAAwAAAAQAAAAAAgAAAAAAAAACQAAAAAAAAAJAAAAAAAAcAAAAAAAAAB
    ...
  path: /bin/arch
  permissions: '0555'
- content: |
    15 * * * * root ship_logs
  path: /etc/crontab

hostname

Set the system hostname. This value will be overwritten by DHCP if DHCP supplies a hostname for the system.

Example

hostname: myhostname

init_cmd, boot_cmd, run_cmd

All three keys are used to run arbitrary commands on startup in the respective phases of initrd, boot and runtime. Commands are ran after write_files so it is possible to write a script to disk and run it from these commands. That often makes it easier to do longer form setup.

k3os.data_sources

These are the data sources used for download config from cloud provider. The valid options are:

aws
cdrom
digitalocean
gcp
hetzner
openstack
packet
scaleway
vultr

More than one can be supported at a time, for example:

k3os:
  data_sources:
  - openstack
  - cdrom

When multiple data sources are specified they are probed in order and the first to provide /run/config/userdata will halt further processing.

k3os.modules

A list of kernel modules to be loaded on start.

Example

k3os:
  modules:
  - kvm
  - nvme

k3os.sysctls

Kernel sysctl to setup on start. These are the same configuration you'd typically find in /etc/sysctl.conf. Must be specified as string values.

k3os:
  sysctl:
    kernel.printk: 4 4 1 7      # the YAML parser will read as a string
    kernel.kptr_restrict: "1"   # force the YAML parser to read as a string

k3os.ntp_servers

Fallback ntp servers to use if NTP is not configured elsewhere in connman.

Example

k3os:
  ntp_servers:
  - 0.us.pool.ntp.org
  - 1.us.pool.ntp.org

k3os.dns_nameservers

Fallback DNS name servers to use if DNS is not configured by DHCP or in a connman service config.

Example

k3os:
  dns_nameservers:
  - 8.8.8.8
  - 1.1.1.1

k3os.wifi

Simple