pyspark change data capture implementation

Question

I have one base table, which is holding the actual data. below is the table structure

id	name	address	age	date
A1	{"fname": "Alex", "lname": "Bhatt"}	{"lane": "Mac Street", "flat": ["24", "26", "27", "29"]}	56	20201128
A2	{"fname": "Bob", "lname": "Natarajan"}	{"lane": "Royd Street", "flat": ["22", "23", "27"], "pin": "123514"}	53	20201123
A1	{"fname": "Alex", "lname": "Bhattacharya"}	{"lane": "Mac Street", "flat": ["24", "26", "27", "29"]}	56	20201228
A2	{"fname": "Bob", "lname": "Natarajan"}	{"lane": "Royd Street", "flat": ["22", "24", "27"], "pin": "123514"}	53	20201228

In the above table for A1 and A2 there is change in data. This changed data summary is captured and provided by another table. The table structure is mentioned below.

id	changed_field	date
A1	name.lname	20201228
A2	address.flat[1]	20201228

From the above 2 table, I have to prepare the final table where the detail of changed data will be captured. Below is the expected table.

id	changed_field	new_value	newdate	old_value	olddate
A1	name.lname	Bhattacharya	20201228	Bhatt	20201128
A2	address.flat[1]	24	20201228	23	20201123

I have tried with spark sql functions get_json_object() but it is not working. Any suggestion will be really helpful

what are the data types of name and address columns? StringType or StructType? — jxc, Dec 29 '20 at 04:49

score 0 · Accepted Answer · edited Dec 30 '20 at 08:36

I think you need to create another json column in order to use get_json_object... see my answer below.

import pyspark.sql.functions as F

result = df1.select(
    'id',
    'date',
    F.to_json(
        F.struct(
            F.from_json('name', 'fname string, lname string').alias('name'),
            F.from_json('address', 'lane string, flat array<string>, pin string').alias('address')
        )
    ).alias('jsoncol')
).join(
    df2.withColumnRenamed('date', 'date2'), 'id'
).withColumn(
    'new_value',
    F.expr("get_json_object(jsoncol, '$.' || changed_field)")
).groupBy('id', 'changed_field').agg(
    F.array_sort(
        F.collect_list(
            F.array('date', 'new_value')
        )
    ).alias('values')
).select(
    'id',
    'changed_field',
    F.col('values')[1][1].alias('new_value'),
    F.col('values')[1][0].alias('newdate'),
    F.col('values')[0][1].alias('old_value'),
    F.col('values')[0][0].alias('olddate')
)

result.show(truncate=False)
+---+---------------+------------+--------+---------+--------+
|id |changed_field  |new_value   |newdate |old_value|olddate |
+---+---------------+------------+--------+---------+--------+
|A1 |name.lname     |Bhattacharya|20201228|Bhatt    |20201128|
|A2 |address.flat[1]|24          |20201228|23       |20201123|
+---+---------------+------------+--------+---------+--------+

pyspark change data capture implementation

1 Answers1