-
Notifications
You must be signed in to change notification settings - Fork 95
Description
I set up a fresh cluster with local kind as management, the hetzner-hcloud-control-planes flavor, following the caph documentation, with a finished result of:
- 3 control plane nodes in hetzner cloud
- 3 worker nodes in hetzner cloud
- 3 baremetal worker nodes
After the first control plane node, hccm was installed (syself fork).
Then after some more time, clusterctl showed that everything is good to go.
For hetzner cloud worker nodes I want volumes support, so I installed the hetzner CSI.
The issue: With defaults, hetzner CSI agents want to start on baremetal worker nodes as well.
The hetzner CSI deployment/daemonset would already provide support and not spawn on baremetal servers if any of these node labels are set, see https://github.com/hetznercloud/csi-driver/blob/main/deploy/kubernetes/hcloud-csi.yml#L172
instance.hetzner.cloud/is-root-server="true"
instance.hetzner.cloud/provided-by="robot"
Neither of these were set. After setting them manually, CSI behaved as expected and everything works fine.
From other clusters with hetzner cloud controller manager I see a lot more labels per default on each server, but I have not checked where these baremetal related labels are actually set.
I think this is not a CAPH issue, but more likely a CAPH documentation issue, and/or I missed some steps.
What I would have expected is either:
- Support for CSI default labels in CAPH/documentation/templates (preferable)
or - Some hint in the documentation that labels for CSI should be set manually when also running baremetal servers
Environment:
- cluster-api-provider-hetzner version: 1.0.7
- Kubernetes version: 1.33.5
- hccm: syself variant 2.0.1