kilo/docs/userspace-wireguard.md
Lucas Servén Marín e7855825cf
docs/userspace-wireguard.md: add details
This commit clarifies a few lines from the userspace doc and notes in
the README that Kilo works with userspace WireGuard.

Signed-off-by: Lucas Servén Marín <lserven@gmail.com>
2021-01-07 13:48:10 +01:00

1.9 KiB

Userspace WireGuard

It is possible to use a userspace implementation of WireGuard with Kilo. This can make sense in cases where

  • not all nodes in a cluster have WireGuard installed; or
  • nodes are effectively immutable and kernel modules cannot be installed.

Homogeneous Clusters

In a homogeneous cluster where no node has the WireGuard kernel module, a userspace WireGuard implementation can be made available by deploying a DaemonSet. This DaemonSet creates a WireGuard interface that Kilo will manage. In order to avoid race conditions, kg needs to be passed the --create-interface=false flag.

An example configuration for a k3s cluster with boringtun can be applied with:

kubectl apply -f https://raw.githubusercontent.com/squat/Kilo/master/manifests/kilo-k3s-userspace.yaml

Note: even if some nodes have the WireGuard kernel module, this configuration will cause all nodes to use the userspace implementation of WireGuard.

Heterogeneous Clusters

In a heterogeneous cluster where some nodes are missing the WireGuard kernel module, a userspace WireGuard implementation can be provided only to the nodes that need it while enabling the other nodes to leverage WireGuard via the kernel module. An example of such a configuration for a k3s cluster can by applied with:

kubectl apply -f https://raw.githubusercontent.com/squat/Kilo/master/manifests/kilo-k3s-userspace-heterogeneous.yaml

This configuration will deploy nkml as a DaemonSet to label all nodes according to the presence of the WireGuard kernel module. It will also create two different DaemonSets with Kilo: kilo without userspace WireGuard and kilo-userspace with boringtun as a sidecar. Note: because Kilo is dependant on nkml, nkml must be run on the host network before CNI is available and requires a kubeconfig in order to access the Kubernetes API.