Along with disruption monitoring via external endpoint we should add in-cluster monitors which run the same checks over:

• service network (kubernetes.default.svc)
• api-int endpoint (via hostnetwork)
• localhosts (on masters only)

These tests should be implemented as deployments with anti-affinity landing on different nodes. Deployments are selected so that the nodes could properly be drained. These deployments are writing to host disk and on restart the pod will pick up existing data. When a special configmap is created the pod will stop collecting disruption data.

External part of the test will create deployments (and necessary RBAC objects) when test is started, create stop configmap when it ends and collect data from the nodes. The test will expose them on intervals chart, so that the data could be used to find the source of disruption