Kubernetes Node Distribution and Availability: Implementation and Testing Notes

Implementation notes

A Kubernetes clusters control plane must be distributed over multiple physical machines, as well as different "failure zones". How these are defined is at the moment up to the CSP. Worker nodes can also be distributed over "failure zones", but this isn't a requirement. Distribution must be shown through labelling, so that users can access these information.

Node distribution metadata is provided through the usage of the labels topology.kubernetes.io/region, topology.kubernetes.io/zone and topology.scs.community/host-id respectively.

At the moment, not all labels are set automatically by most K8s cluster utilities, which incurs additional setup and maintenance costs.

Automated tests

Notes

The test for the SCS K8s Node Distribution and Availability checks if control-plane nodes are distributed over different failure zones (distributed into physical machines, zones and regions) by observing their labels defined by the standard.

Implementation

The script k8s_node_distribution_check.py connects to an existing K8s cluster and checks if a distribution can be detected with the labels set for the nodes of this cluster.

Manual tests

None.