priority order between multiple autoscalable worker node pools ?

[4FunAndProfit]

Hello @vitobotta

(Yes I have another question sorry haha)
Does hetzner-k3s support defining a priority order between multiple autoscalable worker node pools (for example: scale a cheaper instance pool first, and only fall back to more expensive pools if needed because availability )?

More specifically: does the order of worker_node_pools in the YAML configuration matter, or is the only way to influence this behavior through pod scheduling constraints such as nodeAffinity, nodeSelector, or similar mechanisms?

Thanks a lot in advance ! :)

PS : saw this I don't know if it is compatible https://docs.digitalocean.com/products/kubernetes/how-to/autoscale/?utm_source=chatgpt.com#configuring-priority-expander with a configmap

[vitobotta]

Hi, hetzner-k3s sets up the autoscaler but the functionality available for the autoscaling depends on the specific implementation of the cluster autoscaler for the provider, in this case Hetzner. I am not aware of any feature in such implementation that would allow you to define a priority order between multiple autoscalable worker node pools.

[4FunAndProfit]

Sorry @vitobotta to ask again a related question.

From what I understand in the official Cluster Autoscaler documentation, the priority expander seems to be a generic Cluster Autoscaler feature, not something provider-specific:
https://github.com/kubernetes/autoscaler/blob/master/cluster-autoscaler/expander/priority/readme.md

If that’s correct, does it mean that priority-based scale-up should work with the Hetzner Cluster Autoscaler as well, at least in theory?
Or is there a known limitation in the Hetzner provider implementation that prevents the priority expander from being effective in practice?

Thanks again for the project and for clarifying 🙏

[vitobotta]

Hi, I am not sure to be honest because I have heard about this feature for the first time from you. :)

[4FunAndProfit]

Okay :) thanks a lot for the reply! 🙂

[ArgirisGkogkidis]

@4FunAndProfit did you manage to test this? that is a really interesting feature.

[4FunAndProfit]

@ArgirisGkogkidis Unfortunately, I’m not really sure how to properly test this on my side.
If anyone manages to make it work, I’d be super interested in a short write-up explaining how you did it.

It could be a really interesting feature — especially if it helps optimize infrastructure costs and potentially make this tool even more indispensable 🙂