CheckNodePerf is running on non master nodes, when the worker role label is not present. 

in a Vmware cluster create a infra MCP, and label a node as role:infra

vsphere-problem-detector-operator will produce CheckNodePerf alerts and logs like

CheckNodePerf: xxxxxx failed: master node has disk latency of greater than 100ms

https://docs.openshift.com/container-platform/4.10/machine_management/creating-infrastructure-machinesets.html#creating-infra-machines_creating-infrastructure-machinesets

1.
2.
3.

CheckNodePerf: xxxxx failed: master node has disk latency of greater than 100ms

no log entry, and no alert

The code only considers worker and master labels, also very complex nesting of conditions.

https://github.com/openshift/vsphere-problem-detector/blob/ca408db88a70cfa5aefa3128dff971a555994c29/pkg/check/node_perf.go#L133-L143