When the skew and parallel parameters are true simultaneously, the result is less data

When the params of hive.optimize.skewjoin, hive.groupby.skewindata and hive.exec.parallel are true, and exec sql such as 'INSERT... FROM (SUBQUERY UNIONALL ...GROUP BY...) A JOIN/LEFT JOIN A.expression', result data will be reduced. Details of SQL and test data can be found in the attachment