SCD Type 2 Using MERGE in delta

[AndersASorensen]

Whats the best practice when I want to both update AND insert a record into a target table in one operation ensureing atomisity?

newRecord

ID, Product, Color

123, A, Blue

myTable

ID, Product, Color, is_current

123, A, Yellow, TRUE

I basically want to do an atomic operation that gives the following result in myTable

ID, Product, Color, is_current

123, A, Yellow, FALSE

123, A, Blue, TRUE

I basically want to do a .whenMatchedUpdate AND a .whenMatchedInsert in a MERGE statement, but I know thats not supported.

[nilendraFabric]

Hello @AndersASorensen 

The `MERGE` statement is not natively supported in Microsoft Fabric’s Data Warehouse as of now. Users need to rely on PySpark or SQL workarounds for similar functionality

from delta.tables import DeltaTable

# Define source and target
target_table = DeltaTable.forName(spark, "Lakehouse.TargetTable")
source_df = spark.read.table("Lakehouse.SourceTable")

# Perform the merge
target_table.alias("target").merge(
source=source_df.alias("source"),
condition="target.key = source.key"
).whenMatchedUpdateAll().whenNotMatchedInsertAll().execute()

https://aventius.co.uk/2024/03/28/microsoft-fabric-using-pyspark-to-dynamically-merge-data-into-many...

Currently, the `MERGE` statement is not supported in Fabric’s Data Warehouses. As a workaround:
• Use a combination of `ROW_NUMBER()` for deduplication and `INSERT`/`UPDATE` statements to achieve similar functionality:

WITH DeduplicatedSource AS (
SELECT *, ROW_NUMBER() OVER (PARTITION BY ID ORDER BY LOAD_DATE DESC) AS row_num
FROM SourceTable
WHERE row_num = 1
)
-- Merge logic
MERGE INTO TargetTable AS target
USING DeduplicatedSource AS source
ON target.ID = source.ID
WHEN MATCHED THEN UPDATE SET target.Column = source.Column
WHEN NOT MATCHED THEN INSERT (Column) VALUES (source.Column);

Hope this helps 

Thanks

[AndersASorensen]

I am happy to use pyspark, and I want to do it on a lakehouse, not a DW.

However, your .whenMatchedUpdateAll().whenNotMatchedInsertAll().execute() suggestion will not work as this will only update the matched row.

Yes, I would like it to update a is_current column when matched, but it will not insert the record as a new row too.

[nilendraFabric · Accepted solution]

Hello @AndersASorensen 

give it a try

from pyspark.sql.functions import *
from delta.tables import *

# Assuming 'target_table' is your existing Delta table
target_table = DeltaTable.forName(spark, "target_table")

# 'source_df' is your new data
merge_condition = "target.natural_key = source.natural_key"

(target_table.alias("target")
.merge(source_df.alias("source"), merge_condition)
.whenMatchedUpdate(set={
"attribute1": "source.attribute1",
"attribute2": "source.attribute2",
"is_current": lit(True),
"update_date": current_date()
})
.whenNotMatchedInsert(values={
"natural_key": "source.natural_key",
"attribute1": "source.attribute1",
"attribute2": "source.attribute2",
"is_current": lit(True),
"update_date": current_date()
})
.execute())

https://iterationinsights.com/article/implementing-a-hybrid-type-1-and-2-slowly-changing-dimension-i...

Hope this helps.

Please accept the answer if this works. 

Thanks

[AndersASorensen]

Hi again 🙂

The code provided won't work, but the article you linked to had a working solution 🙂

Thanks!

[ricauduro]

This comment saved my life. Thank you @AndersASorensen