Kubernetes has a hell of a learning curve. Once you get your head around it, it can be great but it is a huge lift to learn. For self hosting I’d lean towards docker-compose rather than kubernetes but kubernetes can be nice once you get past the super steep learning curve.

To answer your questions:

1. Any Linux flavor will do the job.
2. No, windows doesn’t really play nice with containers to begin with. Kubernetes is another layer of complexity that will make it tough.
3. Yep, definitely! This is typically done using helm charts. Helm is basically the package manager for kubernetes. There are many helm charts that you can just install into kubernetes and they’ll work for you. You can also roll your own if you have custom applications. Argocd can also be helpful depending on how deep you want to get.
4. I’d start by looking into videos on helm charts. That said, you will need a very strong understanding of docker containers and may need a good understanding of networking.

I would strongly recommend looking into deployment using docker-compose over kubernetes until you understand containers inside and out. While Kubernetes can be nice it akso adds another layer of difficulty. I say this as someone who uses kubernetes daily for work, uses off the shelf helm charts, and writes their own helm charts from scratch.

I can teach you. I have a three node homelab with proxmox and Talos running some basic services. I use fluxcd to deploy. See gitlab.com/devopscoop/talos-template and gitlab.com/devopscoop/fluxcd-template. If you want to to get on a call and screenshare, let me know.

You can get started pretty easily with Podman desktop and kind. Then you have a small self-contained cluster to mess around with, that also functions as a dev environment. Another option, and how I first learned, was with free credits from GCP. I realize this is self-hosted, just providing options for learning…

I’ve always liked Digital Ocean tutorials for getting stuff done. This one shows you how to deploy some things and links out to official documentation for further details: k8s-ops.net/…/getting-started-with-kubernetes-dig…

These are also good free tutorials: www.geeksforgeeks.org/…/kubernetes-tutorial/

Once you are comfortable with kubectl and basic deployments, then look into helm.

You can start with minikube, it’s a one node environment you can run in a vm. Another fairly straightforward kubernetes distribution is k3s, you can set it up on a bunch of nodes running common linux distributions via ansible for example.

I learned how to kubernetes from open online sources and just trying really hard to set up certain apps. Some have kubernetes tutorials or deployment manifests included.

You can also start with helm, which is like a package manager for kubernetes. It can ease you into configuration required for each application.

As for docker, kubernetes uses containers, but those are bundled into pods. It’s not a particularly difficult leap I think. However I recommend trying using podman for a bit first. It’s fairly compatible with dockerfiles and commands, but also allows you to setup and export pods you can then directly use in kubernetes.

It’s not too late for you, you still have a chance. run!

questions.

Do I need a specialized OS for it? No, but they may have additional tools to make it easier.

Could I run this on a Windows server? I’m personally a Linux guy, but a friend who prefers windows server wanted to try it and I thought I’d ask. Yes and ignore every neck beard here who tells you otherwise. Windows + WSL is a good way to have both.

Can I migrate Docker services to a Kube cluster?  How easy is it? Yes, and a moderate amount of work.

Haven’t gotten around to testing Talos enough, but I’m very enthusiastic about the idea. The little I’ve tested it has been great. My reasoning is (from my point of view ofc): maintaining an OS(Linux/windows/what have you) is pain, maintaining Kubernetes is pain. If I can I want to put all my effort into one of them. The regular OSes are made to do anything, Talos is made for one specific thing which hopefully reduces the attack and maintenance surface a lot. You also get a fair amount of handholding with Talos compared to raw Kubernetes.

No idea about the windows server part.

If your docker services is running with docker compose, there’s a tool called kompose.io that will help you. Haven’t tried it, but as most k8s things it probably need some adjustments.

www.youtube.com/watch?v=3I9PkvZ80BQ is an alright overview of the idea k8s. This thing called k3s.io is a nice staying point for testing k8s out

Not sure if the following will help you, but it helped me so I’ll share:

• Kubernetes is isn’t specifically designed for running services. It’s a generalized tool to help you implement a declarative pattern for your compute
• A lot of the hate K8s gets is actually due to the declarative paradigm. There’s still a lot of tasks people don’t consider worthy of describing. In k8s you’re supposed to describe everything

Edit: k8s is awesome in my unpopular? opinion, but yeah. There is a significant learning curve.

A lot of the experience for getting a service up and running is boilerplate. You can checkout an old project of mine to see how a small set of information must be structured to work in k8s: github.com/deifyed/kaex The output will only work for a old version of k8s. Haven’t kept the project updated due to work moving away from k8s

Some tangential advice: don’t use kubernetes for any of your “production”/“essential” homelab services. Feel free to experiment and learn of course, but don’t use it permanently unless you master it and still enjoy it. There are far simpler methods of running containers like Podman Quadlets, which is what I switched to after running K8s for years and made so many headaches go away.