Feature Request: Node Autoscaler

[MatthiasLohr]

With this ticket, I want to suggest to support Node Autoscaling with the hcloud-cloud-controller-manager.

When a Kubernetes cluster runs out of resources (CPU, RAM), new nodes need to be added. It would be awesome if this could be done automatically instead of manually.

hcloud-cloud-controller-manager already supports removing deleted nodes from the Kubernetes cluster, why not supporting also adding nodes, maybe even automatically. Yes, this might be more work compared to removing nodes, but I think all fundamental requirements are already here. hcloud-cloud-controller-manager already has access to the API and therefore can create new servers. Somehow, it's needed to define

• Region
• Operating System
• Machine Type
• SSH Keys
• Name/Name Template

These parameters could be provided by annotations or a ConfigMap to the hcloud-cloud-controller-manager.

Network ID is already provided to the controller (if there is any configured). cloud-init configuration could be used for directly joining the cluster after the new machine is up.

Any thoughts about this?

Best regards
Matthias

[malikkirchner]

This would be greatly appreciated!

[codeagencybe]

I think you can already do this with HPA (Horizontal Pod Autoscaler).
Depending on the metrics you set, it will scale up/down your nodes.

If you have the Hetzner node drivers installed, then the feature is possible to have auto scaling nodes already.

https://kubernetes.io/docs/tasks/run-application/horizontal-pod-autoscale/

This driver can be used with Rancher as example
https://github.com/mxschmitt/ui-driver-hetzner

[niklaskorz]

We (@alugha) have implemented the cloudprovider interface of https://github.com/kubernetes/autoscaler/tree/master/cluster-autoscaler for Hetzner and have been successfully using it for autoscaling on Hetzner Cloud since April. It still has to be adapted to Kubernetes 1.19 (currently using 1.17) but afterwards I see no reason why this couldn't be proposed upstream to be merged into the cluster-autoscaler.
If there's interest, I can publish our current fork of autoscaler here on Github as well before working on a proper pull request.

[MatthiasLohr]

I'm always a friend of Open Source, so why not providing it to the public anyways? 😄

[identw]

I also implemented hetnzer support for autoscaling the cluster. There was no time for a pull request yet, but you can try it https://github.com/identw/autoscaler/tree/hetzner/cluster-autoscaler/cloudprovider/hetzner.
I am new to golang so my code may not be very pretty, however I am using my cluster autoscaler in my clusters, works fine.

[github-actions]

This issue has been marked as stale because it has not had recent activity. The bot will close the issue if no further action occurs.

[LKaemmerling]

Auto-Scaling is unfortunately out of the scope of a Cloud Controller Manager (https://github.com/kubernetes/cloud-provider/blob/master/cloud.go).

[Bessonov]

There is also a pull request: kubernetes/autoscaler#3640