Ansible Role: k3s

Ansible role for installing Rancher Labs k3s ("Lightweight Kubernetes") as either a standalone server or cluster.

Requirements

The host you're running Ansible from requires the following Python dependencies:

• jmespath >= 0.9.0

This role has been tested on Ansible 2.7.0+ against the following Linux Distributions:

• Amazon Linux 2
• CentOS 8
• CentOS 7
• Debian 9
• Debian 10
• Fedora 29
• Fedora 30
• Fedora 31
• Fedora 32
• openSUSE Leap 15
• Ubuntu 18.04 LTS
• Ubuntu 20.04 LTS

Disclaimer

⚠️ May not be suitable for production use!

Rancher Labs is awesome and has released k3s as v1.0.0, however at the time of creating this role I do not have a k3s cluster in production nor am I unlikely to ever have one. Please ensure you practice extreme caution and operational rigor before using this role for any serious workloads.

If you have any problems please create a GitHub issue, I maintain this role in my spare time so I cannot promise a speedy fix delivery.

Role Variables

Group Variables

Below are variables that are set against all of the play hosts for environment consistency.

Variable	Description	Default Value
k3s_cluster_state	State of cluster: installed, started, stopped, restarted, downloaded, uninstalled.	installed
k3s_release_version	Use a specific version of k3s, eg. v0.2.0. Specify false for stable.	false
k3s_build_cluster	When multiple play_hosts are available, attempt to cluster. Read notes below.	true
k3s_github_url	Set the GitHub URL to install k3s from.	https://github.com/rancher/k3s
k3s_install_dir	Installation directory for k3s.	/usr/local/bin
k3s_install_hard_links	Install using hard links rather than symbolic links.	false
k3s_server_manifests_dir	Path for place the k3s_server_manifests_templates.	/var/lib/rancher/k3s/server/manifests
k3s_server_manifests_templates	A list of Auto-Deploying Manifests Templates.	[]
k3s_use_experimental	Allow the use of experimental features in k3s.	false
k3s_use_unsupported_config	Allow the use of unsupported configurations in k3s.	false
k3s_non_root	Install k3s as non-root user. See notes below.	false
k3s_cluster_cidr	Network CIDR to use for pod IPs	10.42.0.0/16
k3s_service_cidr	Network CIDR to use for service IPs	10.43.0.0/16
k3s_control_node_address	Use a specific control node address. IP or FQDN.	NULL
k3s_control_token	Use a specific control token, please read notes below.	NULL
k3s_private_registry	Private registry configuration file (default: "/etc/rancher/k3s/registries.yaml")	NULL
k3s_https_port	HTTPS port listening port.	6443
k3s_use_docker	Use Docker rather than Containerd?	false
k3s_no_flannel	Do not use Flannel	false
k3s_flannel_backend	Flannel backend ('none', 'vxlan', 'ipsec', 'host-gw' or 'wireguard')	vxlan
k3s_no_coredns	Do not use CoreDNS	false
k3s_cluster_dns	Cluster IP for CoreDNS service. Should be in your service-cidr range.	NULL
k3s_cluster_domain	Cluster Domain.	cluster.local
k3s_resolv_conf	Kubelet resolv.conf file	NULL
k3s_no_traefik	Do not use Traefik	false
k3s_no_servicelb	Do not use ServiceLB, necessary for using something like MetalLB.	false
k3s_no_local_storage	Do not use Local Storage	false
k3s_default_local_storage_path	Set Local Storage Path. Specify false for default.	false
k3s_no_metrics_server	Do not deploy metrics server	false
k3s_kube_apiserver_args	Customized flag for kube-apiserver process	[]
k3s_kube_scheduler_args	Customized flag for kube-scheduler process	[]
k3s_kube_controller_manager_args	Customized flag for kube-controller-manager process	[]
k3s_kube_cloud_controller_manager_args	Customized flag for kube-cloud-controller-manager process	[]
k3s_disable_scheduler	Disable Kubernetes default scheduler	false
k3s_disable_cloud_controller	Disable k3s default cloud controller manager.	false
k3s_disable_network_policy	Disable k3s default network policy controller.	false
k3s_write_kubeconfig_mode	Define the file mode from the generated KubeConfig, eg. 644	NULL
k3s_datastore_endpoint	Define the database or etcd cluster endpoint for HA.	NULL
k3s_datastore_cafile	Define the database TLS CA file.	NULL
k3s_datastore_certfile	Define the database TLS Cert file.	NULL
k3s_datastore_keyfile	Define the database TLS Key file.	NULL
k3s_become_for_all	Enable become for all (where value for k3s_become_for_* is NULL	false
k3s_become_for_systemd	Enable become for systemd commands.	NULL
k3s_become_for_install_dir	Enable become for writing to k3s_install_dir.	NULL
k3s_become_for_usr_local_bin	Enable become for writing to /usr/local/bin/.	NULL
k3s_become_for_package_install	Enable become for installing prerequisite packages.	NULL
k3s_become_for_kubectl	Enable become for kubectl commands.	NULL
k3s_become_for_uninstall	Enable become for running uninstall scripts.	NULL
k3s_etcd_datastore	Use Embedded Etcd as the database backend for HA. (EXPERIMENTAL)	false
k3s_secrets_encryption	Use secrets encryption at rest. (EXPERIMENTAL)	false

Important note about k3s_release_version

If you do not set a k3s_release_version the latest version from the stable channel of k3s will be installed. If you are developing against a specific version of k3s you must ensure this is set in your Ansible configuration, eg:

k3s_release_version: v1.16.9+k3s1

It is also possible to install specific K3s "Channels", below are some examples for k3s_release_version:

k3s_release_version: false          # defaults to 'stable' channel
k3s_release_version: stable         # latest 'stable' release
k3s_release_version: testing        # latest 'testing' release
k3s_release_version: v1.18          # latest v1.18 release
k3s_release_version: v1.17-testing  # latest v1.17 testing release

Important node about k3s_install_hard_links

If you are using the system-upgrade-controller you will need to use hard links rather than symbolic links as the controller will not be able to follow symbolic links. This option has been added however is not enabled by default to avoid breaking existing installations.

To enable the use of hard links, ensure k3s_install_hard_links is set to true.

k3s_install_hard_links: true

The result of this can be seen by running the following in k3s_install_dir:

ls -larthi | grep -E 'k3s|ctr|ctl' | grep -vE ".sh$" | sort

Symbolic Links:

[root@node1 bin]# ls -larthi | grep -E 'k3s|ctr|ctl' | grep -vE ".sh$" | sort
3277823 -rwxr-xr-x 1 root root  52M Jul 25 12:50 k3s-v1.18.4+k3s1
3279565 lrwxrwxrwx 1 root root   31 Jul 25 12:52 k3s -> /usr/local/bin/k3s-v1.18.6+k3s1
3279644 -rwxr-xr-x 1 root root  51M Jul 25 12:52 k3s-v1.18.6+k3s1
3280079 lrwxrwxrwx 1 root root   31 Jul 25 12:52 ctr -> /usr/local/bin/k3s-v1.18.6+k3s1
3280080 lrwxrwxrwx 1 root root   31 Jul 25 12:52 crictl -> /usr/local/bin/k3s-v1.18.6+k3s1
3280081 lrwxrwxrwx 1 root root   31 Jul 25 12:52 kubectl -> /usr/local/bin/k3s-v1.18.6+k3s1

Hard Links:

[root@node1 bin]# ls -larthi | grep -E 'k3s|ctr|ctl' | grep -vE ".sh$" | sort
3277823 -rwxr-xr-x 1 root root  52M Jul 25 12:50 k3s-v1.18.4+k3s1
3279644 -rwxr-xr-x 5 root root  51M Jul 25 12:52 crictl
3279644 -rwxr-xr-x 5 root root  51M Jul 25 12:52 ctr
3279644 -rwxr-xr-x 5 root root  51M Jul 25 12:52 k3s
3279644 -rwxr-xr-x 5 root root  51M Jul 25 12:52 k3s-v1.18.6+k3s1
3279644 -rwxr-xr-x 5 root root  51M Jul 25 12:52 kubectl

Important note about k3s_build_cluster

If you set k3s_build_cluster to false, this role will install each play host as a standalone node. An example of when you might be building a large number of IoT devices running K3s. Below is a hypothetical situation where we are to deploy 25 Rasberry Pi devices, each a standalone system and not a cluster of 25 nodes. To do this we'd use a playbook similar to the below:

- hosts: k3s_nodes  # eg. 25 RPi's defined in our inventory.
  vars:
    k3s_build_cluster: false
  roles:
     - xanmanning.k3s

Important note about k3s_non_root

To install k3s as non root you must not use become: true. The intention of this variable is to run a single node development environment. At the time of release v1.0.1, rootless is still experimental.

You must also ensure that you set both k3s_use_experimental and k3s_use_unsupported_config to true.

Additionally k3s_install_dir must be writable by your user.

Important notes about k3s_control_node_address and k3s_control_token

If you set this, you are explicitly specifying the control host that agents should connect to, the value should be an IP address or FQDN.

If the control host is not configured by this role, then you need to also specify the k3s_control_token.

Please note that this may potentially break setting up agents.

Host Variables

Below are variables that are set against specific hosts in your inventory.

Variable	Description	Default Value
k3s_control_node	Define the host as a control plane node, (True/False).	false
k3s_node_name	Define the name of this node.	$(hostname)
k3s_node_id	Define the ID of this node.	NULL
k3s_flannel_interface	Define the flannel proxy interface for this node.	NULL
k3s_bind_address	Define the bind address for this node.	localhost
k3s_node_ip_address	IP Address to advertise for this node.	NULL
k3s_node_external_address	External IP Address to advertise for this node.	NULL
k3s_node_labels	List of node labels.	NULL
k3s_kubelet_args	A list of kubelet args to pass to the server.	[]
k3s_kube_proxy_args	A list of kube proxy args to pass to the server.	[]
k3s_node_taints	List of node taints.	NULL
k3s_node_data_dir	Folder to hold state.	/var/lib/rancher/k3s
k3s_tls_san	Add additional hosname or IP as Subject Alternate Name in the TLS cert.	NULL

Important note about k3s_control_node and High Availability (HA)

By default only one host will be defined as a control node by Ansible, If you do not set a host as a control node, this role will automatically delegate the first play host as a control node (master). This is not suitable for use in a Production workload.

If multiple hosts have k3s_control_node set to true, you must also set k3s_datastore_endpoint as the connection string to a MySQL or PostgreSQL database, or etcd cluster else the play will fail.

If using TLS, the CA, Certificate and Key need to already be available on the play hosts.

See: High Availability with an External DB

It is also possible, though not supported, to run a single K3s master with a k3s_datastore_endpoint defined. As this is not a typically supported configuration you will need to set k3s_use_unsupported_config to true.

Since K3s v1.19.1 it is possible to use Etcd as the backend database, and this is done by setting k3s_etcd_datastore to true. As this is an experimental feature you will also need to set k3s_use_experimental to true.

Important note about k3s_flannel_interface

If you are running k3s on systems with multiple network interfaces, it is necessary to have the flannel interface on a network interface that is routable to the master node(s).

Notes about _args, _labels and _taints variables

Affected variables:

• k3s_kube_apiserver_args
• k3s_kube_scheduler_args
• k3s_kube_controller_manager_args
• k3s_kube_cloud_controller_manager_args
• k3s_kubelet_args
• k3s_kube_proxy_args

These parameters allow for assigning additional args to K3s during runtime. For instance, to use the Azure Cloud Controller, assign the below to the master node's configuration in your host file.

YAML:

k3s_kubelet_args:
  - cloud-provider: external
  - provider-id: azure

Note, when using an external cloud controller as above, ensure that the native k3s cloud controller is disabled by setting the k3s_disable_cloud_controller to true.

Ultimately these variables are lists of key-value pairs that will be iterated on. The below example will output the following:

YAML:

k3s_node_labels:
  - foo: bar
  - hello: world

k3s_node_taints:
  - key1: value1:NoExecute

K3S ARGS:

--node-label foo=bar \
--node-label hello=world \
--node-taint key1=value1:NoExecute

Dependencies

No dependencies on other roles.

Example Playbooks

Example playbook, single master node running v0.10.2:

- hosts: k3s_nodes
  roles:
     - { role: xanmanning.k3s, k3s_release_version: v0.10.2 }

Example playbook, Highly Available running the latest release:

- hosts: k3s_nodes
  vars:
    molecule_is_test: true
    k3s_control_node_address: loadbalancer
    k3s_datastore_endpoint: "postgres://postgres:verybadpass@database:5432/postgres?sslmode=disable"
  pre_tasks:
    - name: Set each node to be a control node
      set_fact:
        k3s_control_node: true
      when: inventory_hostname in ['node2', 'node3']
  roles:
    - role: xanmanning.k3s

License

BSD 3-clause

Author Information

Xan Manning