Description of Problem
The current kill switches (killSwitch and killIfCommands) depend on the Kubernetes API being available. If the API server is unreachable, the reclaim agent cannot PATCH the Node annotation, and the controller cannot act on it. It would be good if the node would kill itself if the k8s control plane becomes unavailable.
Potential Solutions
There are many ways to implement this. The /proc polling service could also poll the k8s API periodically, for instance. You'd want to account for routine network failures.
Description of Problem
The current kill switches (killSwitch and killIfCommands) depend on the Kubernetes API being available. If the API server is unreachable, the reclaim agent cannot PATCH the Node annotation, and the controller cannot act on it. It would be good if the node would kill itself if the k8s control plane becomes unavailable.
Potential Solutions
There are many ways to implement this. The /proc polling service could also poll the k8s API periodically, for instance. You'd want to account for routine network failures.