We stop the per-host probes talking to Kubernetes and run an extra Deployment with one more probe process to collect all information for the cluster, which is less resource-intensive overall. This feature was added at #3274
