Spark skewedPartitionThresholdInBytes not getting enforced

Viewed 103

Relatively new to spark and Curious with the behaviour of PySpark wherein ,

While performing an inner join (According to execution plan :SortMergeJoin ) using PySpark , with following parameters enabled :

spark.sql.adaptive.enabled  true

spark.sql.adaptive.coalescePartitions.enabled    true

spark.sql.adaptive.skewJoin.enabled     true

I see that there are partitions which are greater than spark.sql.adaptive.skewJoin.skewedPartitionThresholdInBytes (256 MB the default value) during job execution. Referring to this

The Median of partition sizes come upto 18MB . spark.sql.adaptive.skewJoin.skewedPartitionFactor is also set to default value 5

However there are 2 partitions existing with sizeable skew :1454.9 MiB and 791.5 MiB respectively .

Isn't the above parameters supposed to ensure that these 2 partitions get split to smaller ones to avoid data skew ?

0 Answers
Related