This was originally implemented for queryable backup, but for sharded clusters, a majority commit point is used as the recovery timestamp. Not all shards may have an oplog entry at the majority commit point, however, and the code expects an oplog entry to be there at the passed in recovery timestamp.