dbench VS local-path-provisioner

Perhaps this might be true for i3.2xlarge or larger machines, but for smaller VMs these were the results I got. My setup was 3 masters - 6 workers (2vCPUs - 4GBs RAM). You can try it yourself with dbench and see what you get.

Deploy local-path-provisioner storage class but it does not support readwritemany so for high availability of your Kubernetes cluster better to use longhorn

I use this one. I'm waiting for the day it's combined with syncthing to sync across all nodes. https://github.com/rancher/local-path-provisioner

So I have this single node k3s cluster. k3s uses local-path (https://github.com/rancher/local-path-provisioner) as default SC that allows one to create dynamic volumes using nodes local storage.

Just create 1 single partition and format it with whatevery filesystem you like. And then use ranchers local-path-provisioner which will create a folder per PV (k3s has this integrated by default).

Tinkering locally with local path provisioner (https://github.com/rancher/local-path-provisioner), I find that I can delete and re-create the pod, and the data persists on disk. However, if I delete the PVC, when I recreate the PVC, a new directory on disk is created.

It is better to use https://github.com/rancher/local-path-provisioner (or similar) for this case which will do PVC on local directories because manually linking PV<>PVC will not work.

Since you only have one physical node anyway, I would just make the cluster a single-node cluster (1 VM) and use local storage on that VM. I’m biased though because this is what I do (I run K3s and use local path provisioner).

Rancher's Local Path Provisioner - From reading, seems to just use HostPath or Local PVs under the hood, but adds dynamic provisoning

With any of those tools, you'd implement a network storage on top of a network storage. I would go with mouting few volumes per node +local storage like (https://github.com/rancher/local-path-provisioner).