KRaft

Kraft is built on top of K3k, a Rancher project to run Kubernetes clusters inside Kubernetes. K3k runs pods on the host cluster with the API server and Kubelets and isolates the different clusters with namespaces, network policies, and more. For Kraft, I am running K3k in shared mode, which uses less resoures at the cost of isolation - it means your guest cluster will inherit access to ingressclasses and storageclasses which are present on the host.

Longhorn is a cloud native block storage solution for Kubernetes, is a Rancher project, also on the CNCF landscape as an incubating project. In the current configuration, Longhorn will create volume replicas on the host nodes, meaning any node could fail and the volume will still work fine. Also, I have both snapshot and backup schedules to Wasabi S3 so, in case I decimate the host cluster by accident, all your data can be recovered. At the end of the day, all of this means that your data is safe and is not at risk of being lost, particularly by hardware failures.
Please keep in mind that you must have good backup practices on your end and not rely entirely on my backups which may or may not work.

KRaft sits on a host cluster with Traefik ingress. This means that when you create an ingress object, it is synced to the host cluster and created by Traefik.

For the time being, KRaft is running on my pre-existing homelab of 3 Dell Optiplex Micros, each with 4 cores, 4 threads, and 32GB of RAM. The nodes are connected over gigabit lines, with a 200/50 Mbps home connection, and I lack a UPS, meaning that powercuts or internet problems will cause noticeable issues.
In the future, the plan is to run KRaft on Talos on an HP Proliant DL 380 Gen8 server with 48 threads and 128GB of RAM. The bonus of running Talos will be security, skipping on having a shell (or many other things). Reducing the attack surface makes my life easier.