Github user KaiXinXiaoLei commented on the issue: https://github.com/apache/spark/pull/20670 @srowen i redescribe the problem. Now i hive a small table `ls` with one row , and a big table `catalog_sales` with One hundred billion rows. And in the big table, the non null value about `cs_order_number` field has one million. Then i join this tables with the query:`select ls.cs_order_number from ls left semi join catalog_sales cs on ls.cs_order_number = cs.cs_order_number`. My job is running, and there has been a data skew. Then i find the null value cause this phenomenon. The join condition is `ls.cs_order_number = cs.cs_order_number`. In the Optimized Logical Plan, the left table has "Filter isnotnull(cs_order_number#1)" action, so i think the right table should have âFilter isnotnullâ action. Then the right table will filter null value firstly , and join with left table secondly. So the data skew will not be caused by null value. Using this idea, my sql runs success.
--- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org