Databricks Pyspark + 如何将数据帧架构作为列插入到数据帧中

Question

I have a function which generates a dataframe:

def getdata():
    schema_1 = StructType([ StructField('path_name', StringType(), True),
                           StructField('age1', IntegerType(), True), 
                           StructField('age2', IntegerType(), True), 
                           StructField('age3', IntegerType(), True)])
    data = [('dbfs/123/sd.zip',1,2,3),('dbfs/123/ab.zip',5,6,7)]
    df = spark.createDataFrame(data,schema_1)
    return df

I need to insert that dataframe schema into a column of another dataframe. The result should be something like:

root
 |-- filename: string (nullable = true)
 |-- parsed: struct (nullable = true)
 |    |-- path_name: string (nullable = true)
 |    |-- age1: integer (nullable = true)
 |    |-- age2: integer (nullable = true)
 |    |-- age3: integer (nullable = true)

I was trying to do it by using a udf:

@udf(schema_1)
def my_udf(schema):
  data = getdata(schema)
  List_of_rows = data.collect()  
  return List_of_rows

And then inserting it into another dataframe I am creating. The whole code I am using is:

from pyspark.sql.types import *
from pyspark.sql.functions import col
import pandas as pd

schema_1 = StructType([ StructField('path_name', StringType(), True),
                           StructField('age1', IntegerType(), True), 
                           StructField('age2', IntegerType(), True), 
                           StructField('age3', IntegerType(), True)])

def getdata(schema_1):    
    data = [('dbfs/123/sd.zip',1,2,3),('dbfs/123/ab.zip',5,6,7)]
    df = spark.createDataFrame(data,schema_1)
    return df


@udf(schema_1)
def my_udf(scheme):
  data = getdata(scheme)
  List_of_rows = data.collect()  
  return List_of_rows

def somefunction(schema_1):    
  pd_df = pd.DataFrame(['path'], columns = ["filename"])  
  return (spark.createDataFrame(pd_df)
          .withColumn('parsed', my_udf(schema_1))
         )

df_2 = somefunction(schema_1)

display(df_2)

However I am getting an error,

Error rendering output: Report error. PicklingError: Could not serialize object: Exception: It appears that you are attempting to reference SparkContext from a broadcast variable, action, or transformation. SparkContext can only be used on the driver, not in code that it run on workers. For more information, see SPARK-5063.

and I also think it is not the best approach. Any idea??

Answer 1

Can you not just create a custom schema? Google for that. Also, see the sample code which will create a (forced) custom schema for you, even if the first rows (headers) are missing, or incorrect.

from  pyspark.sql.functions import input_file_name
from pyspark.sql import SQLContext
from pyspark.sql.types import *
sqlContext = SQLContext(sc)

customSchema = StructType([ \
StructField("asset_id", StringType(), True), \
StructField("price_date", StringType(), True), \

etc., etc., etc., 

StructField("close_price", StringType(), True), \
StructField("filename", StringType(), True)])