Slowly Changing Dimensions implemenation with Databricks Delta Lake

What is Slowly Changing Dimension

Slowly Changing Dimensions (SCD) are dimensions which change over time and in Data Warehuse we need to track the changes of the attributes keep the accuracy of the report.

And typically there are three types of SCD

• Type 1: SCD1, No history preservation
• Type 2: SCD2, Unlimited history preservation and new rows
• Type 3: SCD3, Limited history preservation

For example we have a dataset

ShortName	Fruit	Color	Price
FA	Fiji Apple	Red	3.6
BN	Banana	Yellow	1
GG	Green Grape	Green	2
RG	Red Grape	Red	2

If we change the price of "Fiji Apple" into 3.5, the dataset will be

with SCD1:

ShortName	Fruit	Color	Price
FA	Fiji Apple	Red	3.5
BN	Banana	Yellow	1
GG	Green Grape	Green	2
RG	Red Grape	Red	2

with SCD2:

ShortName	Fruit	Color	Price	is_last
FA	Fiji Apple	Red	3.5	Y
FA	Fiji Apple	Red	3.6	N
BN	Banana	Yellow	1	Y
GG	Green Grape	Green	2	Y
RG	Red Grape	Red	2	Y

with SCD3:

ShortName	Fruit	Color	Price	Color_old	Price_old
FA	Fiji Apple	Red	3.5	Red	3.6
BN	Banana	Yellow	1	NULL	NULL
GG	Green Grape	Green	2	NULL	NULL
RG	Red Grape	Red	2	NULL	NULL

SCD implementation in Databricks

In this repository, there are implementations of SCD1, SCD2 and SCD3 in python and Databricks Delta Lake.

1. SCD1

SCD1(df, target_table_name, target_partition_keys, key_cols, current_time):

Parameters:

• df: source dataframe
• target_table_name: target table name
• target_partition_keys: partition key of the target table
• key_cols: columns of the key for each row
• current_time: current timestamp

Here is the code to show an example of SCD1

from pyspark.sql import functions as F
from pyspark.sql import DataFrame
import datetime

# create sample dataset
df1 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.5)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# prepare parameters
current_time = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
target_partition_keys = ['ShortName']
key_cols = "ShortName,Fruit"
target_table_name_scd1 = 'default.table_scd1'
# call the SCD1 function
SCD1(df1, target_table_name_scd1, target_partition_keys, key_cols, current_time)
# display the result
display(spark.sql(f"select * from {target_table_name_scd1}"))

Change the price of "Fiji Apple" into 3.5 and run SCD1 again

df2 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.6)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# call the SCD1 function again
SCD1(df2, target_table_name_scd1, target_partition_keys, key_cols, current_time)
display(spark.sql(f"select * from {target_table_name_scd1}"))

1. SCD2

SCD2(df, target_table_name, target_partition_keys, key_cols, current_time):

Parameters:

• df: source dataframe
• target_table_name: target table name
• target_partition_keys: partition key of the target table
• key_cols: columns of the key for each row
• current_time: current timestamp

Here is the code to show an example of SCD2

from pyspark.sql import functions as F
from pyspark.sql import DataFrame
import datetime

# create sample dataset
df1 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.5)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# prepare parameters
current_time = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
target_partition_keys = ['ShortName']
key_cols = "ShortName,Fruit"
target_table_name_scd1 = 'default.table_scd2'
# call the SCD1 function
SCD2(df1, target_table_name_scd1, target_partition_keys, key_cols, current_time)
# display the result
display(spark.sql(f"select * from {target_table_name_scd2}"))

Change the price of "Fiji Apple" into 3.5 and run SCD2 again

df2 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.6)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# call the SCD1 function again
SCD2(df2, target_table_name_scd2, target_partition_keys, key_cols, current_time)
display(spark.sql(f"select * from {target_table_name_scd2}"))

1. SCD3

SCD3(df, target_table_name, target_partition_keys, key_cols, current_time):

Parameters:

• df: source dataframe
• target_table_name: target table name
• target_partition_keys: partition key of the target table
• key_cols: columns of the key for each row
• current_time: current timestamp

Here is the code to show an example of SCD3

from pyspark.sql import functions as F
from pyspark.sql import DataFrame
import datetime

# create sample dataset
df1 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.5)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# prepare parameters
current_time = datetime.datetime.now().strftime("%Y-%m-%dT%H:%M:%S")
target_partition_keys = ['ShortName']
key_cols = "ShortName,Fruit"
target_table_name_scd1 = 'default.table_scd3'
# call the SCD1 function
SCD3(df1, target_table_name_scd3, target_partition_keys, key_cols, current_time)
# display the result
display(spark.sql(f"select * from {target_table_name_scd3}"))

Change the price of "Fiji Apple" into 3.5 and run SCD3 again

df2 = spark.createDataFrame([('FA', 'Fiji Apple', 'Red', 3.6)
                           ,('BN', 'Banana', 'Yellow', 1.0)
                           ,('GG', 'Green Grape', 'Green', 2.0)
                           ,('RG', 'Red Grape', 'Red', 2.0)], 
                           ['ShortName','Fruit', 'Color', 'Price'])
# call the SCD1 function again
SCD3(df2, target_table_name_scd3, target_partition_keys, key_cols, current_time)
display(spark.sql(f"select * from {target_table_name_scd3}"))