This is my Kubernetes Homelab, managed with Kluctl, of which I'm also the main developer and maintainer.
It's still a very young Homelab and thus doesn't do a lot. It basically just sets up what is usually needed (e.g. Cilium, cert-manager, ingress-nginx, metallb, ...) and then deploys a small set of applications I need at home. This will surely grow over tine.
My Homelab is based on multiple Raspberry Pis that act as the K8s control plane and worker nodes. I plan to enlarge this cluster when necessary with NUCs or comparable hardware to also get some x64 nodes. Storage is simply solved via a Synology DS and NFS volumes.
Most K8s Homelabs rely on Flux or ArgoCD to implement a GitOps style workflow. I started to work on Kluctl a few years ago as I did not feel that the existing solutions were suitable for my needs and workflows at work. I was not aware of Flux and ArgoCD and thus did not follow the development of the two major GitOps solutions, which resulted in Kluctl solving the same problems with a completely different approach and point of view.
I believe that the way Kluctl solves complex deployments and configuration management fits very well for many other people. This Homelab can hopefully show a very practical example of how Kluctl solves things and thus motivate people to give it a try.
Next, I'll try to give a short list of main differences between Flux and Kluctl based Homelabs.
Kluctl allows to implement GitOps style workflows while still allowing to revert to non-GitOps workflows whenever needed. This is possible because Kluctl deployment projects do not rely on CRDs and thus also don't need a controller to perform a deployment.
This in turn allows to perform deployments from your local machine, which is handy when you need to move fast (no more push+pray iterations) and test out changes. You can also perform dry-run deploys/diffs locally, which works even better in combination with GitOps. When you're done with your changes and feel confident enough with the state of the deployment, committing and pushing will then allow GitOps to take over. GitOps can either be solved via the kluctl-controller (recommended) or via CI/CD pipelines.
There is no need to use another tool (e.g. Ansible) to bootstrap a cluster with a CNI solution or any other base component. Kluctl can easily install Cilium (or any other CNI) in the same deployment as any other component that is needed for any upcoming deployments. You can really start using Kluctl on a completely naked cluster.
Kluctl is not based on CRDs and instead relies on a hierarchical project structure. This allows for very simple and easy dependency/order management in a declarative and predictable way.
Check base/deployment.yaml for example. Each deployment item is processed in parallel until a barrier forces Kluctl to wait for completion of previous deployments. It then continues with the next deployment items until the next barrier appears. This way, it's easy to deploy namespaces before anything else, CRDs before CRs, and so on...
Kluctl uses templating and variable sources to glue together individual deployment items. Templating can be used absolutely everywhere, meaning that everything that configuration can be used everywhere to control the deployment.
I still recommend to use templating at a minimal level, so that it doesn't get too complex. In this Homelab deployment for example, I mostly use templating to distribute network related configuration.
I still plan to add a few things here and there. The most important TODOs:
- Actually enable GitOps in this repo. This means I need to deploy the kluctl-controller and then take over deployments of this repo.
- Introduce Renovate to update helm-chart.yaml files. I have a working configuration already, it just needs to be incorporated here.
- Move away from Ansible (I hate Ansible...) and find a way to use Talos, PXE boot, and NFS for the rootfs
- More applications :)
The Ansible/K3s deployment and Taskfiles are based on https://github.com/onedr0p/flux-cluster-template, which saved me quite some time with my Raspberry PI + K3s based Homelab.