Generally speaking atClusterTime readConcern is sent via mongos, which will also include $clusterTime metadata to advance the clock. If, however, an atClusterTime readConcern was received without $clusterTime metadata, or with a $clusterTime less than the atClusterTime value, then the no-op write performed by the read concern machinery might fail to advance the clock, and the read can block forever waiting for the cluster time to advance. Instead of hanging forever, the read should fail.