Hello all, my regular inner join in Flink SQL (v1.20.3) creates heavy data skew, although the input already contains evenly distributed partitions after I salted the keys with a random int ranging from (0, operatorPallelism). Interestingly when i increase the salting range to eg (0, operatorParallelism * 8) it gets more evenly distributed. I thought this should have no effect as the regular join anyway already assigns every salt key to a new partition / subtask in a round-robin fashion or so. Can someone explain how Flink assigns keys to partitions in a regular inner join? I tried to find it in the source code, but can't find the section where partitioning decides where a join key goes into. Thanks :)