stork VS k8s-storage-provider-benchmarks

What you're looking for is a concept known as fencing. I.e when the node dies the pod is fenced (deleted) to allow rescheduling. I believe that the Stork scheduler can do this. I don't know exactly how Stork handles this but for OnDat (the artist formerly known as StorageOS) we have implemented a similar feature with added safety measures as our nodes gossip so we know that the node is really dead.

Colocated makes a lot of sense as well, because real workloads will almost certainly be as close as they can get to storage. There's even stork which tries to help the scheduler make it so (though you can do it other simpler ways).

Thank you for this! I also use ansible to manage my nodes (very old here)so this is going right in the playbook.

k0s is fantastic -- I used it recently in some storage provider testing (Code is on GitHub, includes ansible code that sets up k0s v1.12.0 on Ubuntu 20.04).

Definitely bookmarked that repo though, cool to see other peoples' setups -- my setup is not open source but I open source parts of it when I make repos to go with blogposts, thanks for the link.

Feel free to email me or file issues directly on the repo, I've already added a few for performance tuning ideas after talking to the folks at OpenEBS.

You can watch the action happen in realtime on the GitLab repo.