Network Routes are not properly cleaned up by hcloud-cloud-controller-manager since HCCM v1.20.0.
This causes issues when users cycle through their nodes (with autoscaler) and they have a lot of stale routes laying around. Adding new nodes then fails because they have already reached the routes limit (100).
HCCM only cleans up routes that are inside its configured --cluster-cidr.
In HCCM v1.20.0 we changed our pod spec from using command to args. Your patches (templates/ccm.yaml.tpl) still try to override the command, and are then ignored because the ccm-networks.yaml has the default --cluster-cidr in the args.
Description
Network Routes are not properly cleaned up by hcloud-cloud-controller-manager since HCCM v1.20.0.
This causes issues when users cycle through their nodes (with autoscaler) and they have a lot of stale routes laying around. Adding new nodes then fails because they have already reached the routes limit (100).
HCCM only cleans up routes that are inside its configured
--cluster-cidr
.In HCCM v1.20.0 we changed our pod spec from using
command
toargs
. Your patches (templates/ccm.yaml.tpl
) still try to override thecommand
, and are then ignored because theccm-networks.yaml
has the default--cluster-cidr
in theargs
.This was originally reported in https://github.com/hetznercloud/hcloud-cloud-controller-manager/issues/734
Kube.tf file
Screenshots
No response
Platform
Linux