Warning

This is just a toy provider that I wrote based on other native providers I found (mainly azure at the time). It is not maintained since I personally don't use k3s anymore.

I now think this provider is way to complicated and something that could be achieved using the command provider as I did in another project.

If you want to build a provider, there's now also a new option that this repo doesn't use.

k3s Pulumi Provider

This is a side project and not intended for production usage. That's why currently only single node setups are supported and there are no released binaries.

To get more information on what decisions have been made for this project have a look at the architecture decision records in the docs.

pulumi-k3s closes the gap between existing Pulumi providers for IaaS (VMs) and the Kubernetes provider. With pulumi-k3s it is possible to define all your required infrastructure as well as Kubernetes resources in one pulumi stack with proper dependency tracking while provisioning your whole landscape.

An example of using this provider's Cluster resource can be found in the examples.

To get a quick preview, here's a minimal example:

const cluster = new k3s.Cluster("mycluster", {
    // currently only one master node is supported in the following
    // list. To not break compatibility with future releases it's already
    // designed to have multiple master nodes.
    masterNodes: [
        {
            host: cfg.require("master_ip"),
            user: "ubuntu",
            privateKey: cfg.requireSecret("private_key"),
        }
    ],
    versionConfig: {
        version: "v1.22.3+k3s1"
    }
});

// cluster.kubeconfig is the output that contains the kubeconfig used
// for the Pulumi Kubernetes provider.
...

Features

As mentioned pulumi-k3s is a side project to support my infrastructure setup which currently only consists of a single node cluster. Therefore, neither multiagent nor multi master setups are currently supported.

Still, this provider already has some nice features including:

• version bumps either automatically using one of the k3s channels or manually by bumping the version and reapplying the Pulumi stack.
• support for the gVisor runtime class
• seamless integration with the Pulumi Kubernetes provider
• custom installation arguments for the nodes to for example disable the default Traefik ingress controller

Installation

Currently, this provider is not optimized for distribution. Therefore, if you want to use it, you need to follow the instructions defined in the Development section and build the provider as well as the needed SDK for you preferred language from the main branch and use that.

Development

Most of the code for the provider implementation is in pkg/provider/provider.go.

A code generator is available which generates SDKs in TypeScript, Python, Go and .NET which are also checked in to the sdk folder. The SDKs are generated from a schema in provider/cmd/pulumi-resource-k3s/schema.json. This file should be kept aligned with the resources, functions and types supported by the provider implementation.

Note that the generated provider plugin (pulumi-resource-k3s) must be on your PATH to be used by Pulumi deployments. If creating a provider for distribution to other users, you should ensure they install this plugin to their PATH.

Pre-requisites

Install the pulumictl CLI from the releases page or follow the installation instructions.

NB: Usage of pulumictl is optional. If not using it, hard code the version in the Makefile of when building explicitly pass version as VERSION=0.0.1 make build

Build and Test

# build and install the resource provider plugin
$ make build install

# test
$ cd examples/simple
$ yarn link @pulumi/k3s
$ yarn install
$ pulumi stack init test
$ pulumi up