When ParallelIndexScan is used, individual IndexScan operators work together under an Exchange operator to do an index scan in parallel. As each IndexScan operator requires rows, it takes the next unprocessed leaf page and returns rows from the page, one at a time. In this way, pages are divided between the IndexScan operators to achieve parallel processing. Regardless of how the pages are distributed among the IndexScan operators, all of the rows are visited.