加入 pyspark 中多列的通用合并

Question

I have to merge many spark DataFrames. After the merge, I want to perform a coalesce between multiple columns with the same names.

I was able to create a minimal example following this question .

However, I need a more generic piece of code to support: a set of variables to coalesce (in the example set_vars = set(('var1','var2')) ), and multiple join keys (in the example join_keys = set(('id')) ).

Is there a less verbose (more generic) way to obtain this result in pyspark ?

df1 =  spark.createDataFrame([ 
        ( 1, None ,  "aa"),
        ( 2 , "a", None ),
        ( 3 , "b",  None),
        ( 4 , "h",  None),],
        "id int, var1 string, var2 string",
       )

df2 =  spark.createDataFrame([ 
        ( 1, "f" ,  "Ba"),
        ( 2 , "a", "bb" ),
        ( 3 , "b",  None),],
        "id int, var1 string, var2 string",
       )

df1 = df1.alias("df1")
df2 = df2.alias("df2")

df3 = df1.join(df2, df1.id == df2.id, how='left').withColumn("var1_", coalesce("df1.var1", "df2.var1")).drop("var1").withColumnRenamed("var1_", "var1").withColumn("var2_", coalesce("df1.var2", "df2.var2")).drop("var2").withColumnRenamed("var2_", "var2")

Answer 1

We can avoid duplicate columns by passing columns as a list to join method instead of writing joining condition, refer this link . But here there are some common columns which are not required for joining condition. we can use for loop to generalize your code.

spark = SparkSession.builder.master("local[*]").getOrCreate()

df1 =  spark.createDataFrame([
        ( 1, None ,  "aa"),
        ( 2 , "a", None ),
        ( 3 , "b",  None),
        ( 4 , "h",  None),],
        "id int, var1 string, var2 string",
       )

df2 =  spark.createDataFrame([
        ( 1, "f" ,  "Ba"),
        ( 2 , "a", "bb" ),
        ( 3 , "b",  None),],
        "id int, var1 string, var2 string",
       )

df1 = df1.alias("df1")
df2 = df2.alias("df2")

key_columns = ["id"]
# Get common columns between 2 dataframes excluding columns-
# -which are being used in joining conditions
other_common_columns = set(df1.columns).intersection(set(df2.columns))\
.difference(set(key_columns))

outputDF = df1.join(df2, key_columns, how='left')

for i in other_common_columns:
    outputDF = outputDF.withColumn(f"{i}_", coalesce(f"df1.{i}", f"df2.{i}"))\
.drop(i).withColumnRenamed(f"{i}_", i)

outputDF.show()

+---+----+----+
| id|var2|var1|
+---+----+----+
|  1|  aa|   f|
|  3|null|   b|
|  4|null|   h|
|  2|  bb|   a|
+---+----+----+