How to convert complex SQL query to spark-dataframe using python or Scala

Question

I have done one transformation using sqlcontext in spark but same query I want to write using Spark Data frame only . This query include join operation plus case statement of SQL. The sql query written as below:

refereshLandingData=spark.sql( "select a.Sale_ID, a.Product_ID,"
                           "CASE "
                           "WHEN (a.Quantity_Sold IS NULL) THEN b.Quantity_Sold "
                           "ELSE a.Quantity_Sold "
                           "END AS Quantity_Sold, "
                           "CASE "
                           "WHEN (a.Vendor_ID IS NULL) THEN b.Vendor_ID "
                           "ELSE a.Vendor_ID "
                           "END AS Vendor_ID, "
                           "a.Sale_Date, a.Sale_Amount, a.Sale_Currency "
                           "from landingData a left outer join preHoldData b on a.Sale_ID = b.Sale_ID" )

now I want equvalent code in spark dataframe in both scala and python. I have tried some code but its
not working .my tried code is as follow:

joinDf=landingData.join(preHoldData,landingData['Sale_ID']==preHoldData['Sale_ID'],'left_outer')

joinDf.withColumn\
('QuantitySold',pf.when(pf.col(landingData('Quantity_Sold')).isNull(),pf.col(preHoldData('Quantity_Sold')))
.otherwise(pf.when(pf.col(preHoldData('Quantity_Sold')).isNull())),
 pf.col(landingData('Quantity_Sold'))).show()

In the above code joining done perfectly but case condition not working. I am getting--> TypeError: 'DataFrame' object is not callable I am using spark 2.3.2 version and python 3.7 and similarly scala 2.11 in case of spark-scala Please anyone suggest me any equivalent code or guidence !

Answer 1

Here's a scala solution : Assuming landingData and preHoldData are your dataframes

 val landingDataDf = landingData.withColumnRenamed("Quantity_Sold","Quantity_Sold_ld")
 val preHoldDataDf = preHoldData.withColumnRenamed("Quantity_Sold","Quantity_Sold_phd")

 val joinDf = landingDataDf.join(preHoldDataDf, Seq("Sale_ID"))


 joinDf
 .withColumn("Quantity_Sold",
    when(col("Quantity_Sold_ld").isNull , col("Quantity_Sold_phd")).otherwise(col("Quantity_Sold_ld"))
 ). drop("Quantity_Sold_ld","Quantity_Sold_phd")

You can do the same way for Vendor_id

The problem with your code is, you cannot reference the other/old dataframe names in withColumn operation. It has to be from the dataframe that you are operating.

Answer 2

Below code will work on scala & for python you might tune little bit.

val preHoldData = spark.table("preHoldData").alias("a")
val landingData = spark.table("landingData").alias("b")

landingData.join(preHoldData,Seq("Sale_ID"),"leftouter")
.withColumn("Quantity_Sold",when(col("a.Quantity_Sold").isNull, col("b.Quantity_Sold")).otherwise(col("a.Quantity_Sold")))
.withColumn("Vendor_ID",when(col("a.Vendor_ID").isNull, col("b.Vendor_ID")).otherwise(col("a.Vendor_ID")))
.select(col("a.Sale_ID"),col("a.Product_ID"),col("Quantity_Sold"),col("Vendor_ID"),col("a.Sale_Date"),col("a.Sale_Amount"),col("a.Sale_Currency"))