Store string in a column as nested JSON to a JSON file - Pyspark

Mister Spurious :

I have a pyspark dataframe, this is what it looks like

+------------------------------------+-------------------+-------------+--------------------------------+---------+
|member_uuid                         |Timestamp          |updated      |member_id                       |easy_id  |
+------------------------------------+-------------------+-------------+--------------------------------+---------+
|027130fe-584d-4d8e-9fb0-b87c984a0c20|2020-02-11 19:15:32|password_hash|ajuypjtnlzmk4na047cgav27jma6_STG|993269700|

I transformed the above dataframe to this,

 +---------+---------+-------------------------------------------------------------------------------------------------------------------------------------------------+-------------------+
|attribute|operation|params                                                                                                                                           |timestamp          |
+---------+---------+-------------------------------------------------------------------------------------------------------------------------------------------------+-------------------+
|profile  |UPDATE   |{"member_uuid":"027130fe-584d-4d8e-9fb0-b87c984a0c20","member_id":"ajuypjtnlzmk4na047cgav27jma6_STG","easy_id":993269700,"field":"password_hash"}|2020-02-11 19:15:32|

Using the following code,

ll = ['member_uuid', 'member_id', 'easy_id', 'field']
df = df.withColumn('timestamp', col('Timestamp')).withColumn('attribute', lit('profile')).withColumn('operation', lit(col_name)) \
                    .withColumn('field', col('updated')).withColumn('params', F.to_json(struct([x for x in ll])))
    df = df.select('attribute', 'operation', 'params', 'timestamp')

I have save this dataframe df to a text file after converting it to JSON. I tried using the following code to do the same,

df_final.toJSON().coalesce(1).saveAsTextFile('file')

The file contains,

{"attribute":"profile","operation":"UPDATE","params":"{\"member_uuid\":\"027130fe-584d-4d8e-9fb0-b87c984a0c20\",\"member_id\":\"ajuypjtnlzmk4na047cgav27jma6_STG\",\"easy_id\":993269700,\"field\":\"password_hash\"}","timestamp":"2020-02-11T19:15:32.000Z"}

I want it to save in this format,

{"attribute":"profile","operation":"UPDATE","params":{"member_uuid":"027130fe-584d-4d8e-9fb0-b87c984a0c20","member_id":"ajuypjtnlzmk4na047cgav27jma6_STG","easy_id":993269700,"field":"password_hash"},"timestamp":"2020-02-11T19:15:32.000Z"}

to_json saves the value in the params columns as a string, is there a way to keep the json context here so I can save it as the desired output?

Shu :

Don't use to_json to create params column in dataframe.

The trick here is just create struct and write to the file (using .saveAsTextFile (or) .write.json()) Spark will create JSON for the Struct field.
if we already created json object and writing in json format Spark will add \ to escape the quotes already exists in Json string.

Example:

from pyspark.sql.functions import *

#sample data
df=spark.createDataFrame([("027130fe-584d-4d8e-9fb0-b87c984a0c20","2020-02-11 19:15:32","password_hash","ajuypjtnlzmk4na047cgav27jma6_STG","993269700")],["member_uuid","Timestamp","updated","member_id","easy_id"])

df1=df.withColumn("attribute",lit("profile")).withColumn("operation",lit("UPDATE"))

df1.selectExpr("struct(member_uuid,member_id,easy_id) as params","attribute","operation","timestamp").write.format("json").mode("overwrite").save("<path>")

#{"params":{"member_uuid":"027130fe-584d-4d8e-9fb0-b87c984a0c20","member_id":"ajuypjtnlzmk4na047cgav27jma6_STG","easy_id":"993269700"},"attribute":"profile","operation":"UPDATE","timestamp":"2020-02-11 19:15:32"}

df1.selectExpr("struct(member_uuid,member_id,easy_id) as params","attribute","operation","timestamp").toJSON().saveAsTextFile("<path>")

#{"params":{"member_uuid":"027130fe-584d-4d8e-9fb0-b87c984a0c20","member_id":"ajuypjtnlzmk4na047cgav27jma6_STG","easy_id":"993269700"},"attribute":"profile","operation":"UPDATE","timestamp":"2020-02-11 19:15:32"}