hetzner-k3s delete does not remove worker pool placement groups

[shadshar]

Hi! When running hetzner-k3s delete, it uses the wrong names when it refers to the names of the worker pool placement groups that were created. On creation, the placement groups get named name-X, where X is a number. That number is not being used during the deletion process, so the placement groups remain and need to be deleted by hand using the cloud console.

So for example, with a config snippet as follows:

worker_node_pools:
- name: small-static-1
  instance_type: cax11
  instance_count: 2
  location: nbg1
  labels:
    - key: purpose
      value: k3s-worker
- name: small-static-2
  instance_type: cax11
  instance_count: 2
  location: nbg1
  labels:
    - key: purpose
      value: k3s-worker

During creation, the output confirms the creation like this:

Creating placement group k3s-playground-masters...done.
Creating placement group k3s-playground-small-static-1-1...done.
Creating placement group k3s-playground-small-static-2-1...done.

And deletion fails like this:

Deleting placement group k3s-playground-masters...done.
Placement group k3s-playground-small-static-1 does not exist, skipping.
Placement group k3s-playground-small-static-2 does not exist, skipping.

[vitobotta]

Hi, yeah I'm aware of this and it's on my list of things to fix for the next release. Placement groups don't cost money nor cause any problems so I haven't given this higher priority.

[vitobotta]

Ref. project card https://github.com/users/vitobotta/projects/4/views/1?filterQuery=placement&pane=issue&itemId=59546489

[peterdieleman]

Not sure this is the right place to comment as this is about placement groups, but: I have set up a project and ran create/delete repeatedly (very cool stuff :) ). I also went through the README and found:

NOTE: at the moment instances created by the cluster autoscaler, as well as load balancers and persistent volumes created by deploying your applications must be deleted manually. This may be addressed in a future release.

I would actually say that not deleting the LB can be quite a nice behaviour. When combining it together with something like Cloudflare DNS it avoids reconfiguring the IP in CF. I guess the default should be to delete it as people may run into unexpected costs, but perhaps this behaviour (not deleting the LB) could be supported through a flag.

[vitobotta]

Not sure this is the right place to comment as this is about placement groups, but: I have set up a project and ran create/delete repeatedly (very cool stuff :) ). I also went through the README and found:

NOTE: at the moment instances created by the cluster autoscaler, as well as load balancers and persistent volumes created by deploying your applications must be deleted manually. This may be addressed in a future release.

I would actually say that not deleting the LB can be quite a nice behaviour. When combining it together with something like Cloudflare DNS it avoids reconfiguring the IP in CF. I guess the default should be to delete it as people may run into unexpected costs, but perhaps this behaviour (not deleting the LB) could be supported through a flag.

The README doesn't expand on this but the idea is to ask the user for a confirmation before deleting load balancers and persistent volumes, so you can choose :)