when creating a new MCP with MCselector specified by matchExpressions, it seems required that the value `worker` be part of the valid values of the label "machineconfiguration.openshift.io/role". When that is created even with no nodes with the specified label in nodeSelector, this creates a new MC and it triggers reboot on the worker nodes. 

seen 4.18, but other versions were not tested    

    always on a fresh OCP. reapplying the reproduction steps with completely new MC and MCP doesn't trigger the reboot after the first time. 

    1.on a vanilla fresh OCP cluster that has worker pool create an mcp like below that no node is found to belong to it && no machine config (no node with node label as below) - so completly new mcp:
apiVersion: machineconfiguration.openshift.io/v1
kind: MachineConfigPool
metadata:
  name: my-test
  labels:
    machineconfiguration.openshift.io/role: my-test
spec:
  machineConfigSelector:
    matchExpressions:
      - {
           key: machineconfiguration.openshift.io/role,
           operator: In,
           values: [worker,my-test],
        }
  nodeSelector:
    matchLabels:
      node-role.kubernetes.io/my-test: ""

     2.This creates an empty mcp, and a new MC but also triggers a reboot on the worker nodes. 

    worker nodes are rebooted and rendered MC of the worker MCP is changed

the mcp has 0 machine count so it is anticipaed to have no reboots on any node + no change in the rendered MC of MCP worker.

deleting the added MCP triggers again reboot on the worker nodes
And doesn't remove the corresponding MC, so if the MCP with the same MCselector is created again no reboot would happen on any node. Not sure if this is expected to keep an old MC.