To drive a two phase commit, TransactionCoordinators target shards using the replica set monitors in the ShardRegistry. The TransactionCoordinator does not update these replica set monitors with the response from the targeted host, so if the error is retryable and indicates the targeted host is no longer a primary, e.g. a stepdown error, the TransactionCoordinator will re-target the same node until the corresponding replica set monitor refreshes for some other reason. Instead, the TransactionCoordinator should update the appropriate RSM for each response it receives, like in ShardRemote::_runCommand.