If a node is disconnected from the cluster, a user is able to navigate to the disconnected node and stop processors/controller services. Upon re-joining the cluster, the node should inherit the cluster's run state for all components. This works for processors but appears not to enable controller services as expected. To duplicate, we can create a simple flow, as attached in a template.
Disconnect one node from the cluster. Navigate to the disconnected node, disable the CSVReader controller service. Now re-join back to the cluster. When the node rejoins, instead of the Controller Service being enabled, the service is still disabled on this node, and the ConvertRecord processor is invalid as a result. If you navigate to the Controller Services for the Process Group, though, it shows that the service is enabled. As a result, if the ConvertRecord processor is already running when re-joining, it will show as running but still be invalid on one node.