This is a Helm chart to deploy k3s controlplanes to Kubernetes clusters.
- Loadbalancer with fixed IP
- Persistent Volume for k3s data (Kubeconfig, Certificates, etc)
- Agent-less Mode: controlplane "Nodes" (Pods in our case) do not run workloads and don't count as "Node" in the Cluster
- Fully compliant with Cilium + Longhorn
- Currently a cluster can only consist of a single controlplane pod
- Currently the only valid datasource is sqlite on a local volume
- Currently the final kubeconfig needs to be copied manually from within the container
helm install k3k oci://cargo.ayedo.cloud/library/k3k
The most relevant configuration needs to be done in the loadbalancer and k3s sections of the values. If loadbalancer.enabled is false, the Kubernetes API will not be accessible from outside the host cluster. 99% of the time this is not what you desire.
If the loadbalancer is enabled, an IP Address must be given in loadbalancer.ip. This must be an IP that can be assigned to the Loadbalancer. It's necessary to know this IP in advance so k3s can adjust its TLS certificate and Advertise Address accordingly.
A loadbalancer.className can be given if multiple loadbalancers are availanble in your cluster.
loadbalancer:
enabled: true
class_name: ""
ip: "1.2.3.4"
k3s:
agent_token: "secret"
cluster_cidr: "10.42.0.0/16"
cluster_dns: "10.43.0.10"
cluster_domain: "cluster.local"
data_dir: /var/lib/rancher/k3s
debug: false
disable:
agent: true
flannel: true
helm_controller: true
network_policy: true
traefik: true
localstorage: true
servicelb: true
kube_proxy: true
cloud_controller: false
egress_selector_mode: "pod"
kubeconfig:
filename: kubeconfig.yaml
mode: "0644"
log_level: "0"
service_cidr: "10.43.0.0/16"
service_node_port_range: "30000-32767"
token: "secret"