In https://jira.mongodb.org/browse/BF-41997 we have the case of a CSRS Primary which fails a retryable command (setFCV) due to a step-down request. If the router is quick enough to retry the command before it completes the transition, it can get get into a window where the node is still transitioning.

The node will accept the request, and complete the request. Because of this, the call to waitForCoordinatorsofGivenOfcvToComplete (called by the Config Server as part of the Shard Protocol) will never complete.