topic Advice Needed: Building a Data Pipeline in Microsoft Fabric for Monthly Transactional Data in Pipelines

Advice Needed: Building a Data Pipeline in Microsoft Fabric for Monthly Transactional Data

HamidBee — Sat, 02 Nov 2024 09:11:20 GMT

Hello everyone,

I’m currently working on a data pipeline project in Microsoft Fabric that involves processing transactional data emailed to me monthly, and I’d be grateful for any advice on the best approach to structuring this pipeline.

Project Details:

Data Source: Each month, I receive a non-normalized Excel file containing transaction data via email.
Objective: The goal is to automate the ingestion, transformation, and loading process to organize this data into a star schema within a data warehouse, preparing it for analytics and reporting.
End Use: Once processed, the data should be optimized for fast querying and analysis.

Current Strategy:

Ingestion: Using a Copy Activity to move the Excel data from the Fabric Lakehouse folder into a staging table within the data warehouse.
Transformation: Considering both Stored Procedure and Notebook Activities for data transformations. The process includes cleansing, deduplication, and mapping to final dimension and fact tables.
Normalization: After transformations, I aim to load the data into a normalized schema, with tables like DimLocation, DimTransactionType, and FactTransactions.

Questions:

Stored Procedures vs. Notebooks for Transformations: Since some transformations are complex, would you recommend prioritizing Notebooks (e.g., PySpark) or Stored Procedures? Or a combination of both?
Monthly File Handling: Given the monthly data drop, is it more efficient to use dynamic pipeline parameters for each file, or should I set up new pipelines as files arrive?
Performance Optimization: Any suggestions for optimizing performance, especially for handling large datasets and ensuring transformations remain cost-effective?

Thanks in advance for any insights or experiences you can share.

Re: Advice Needed: Building a Data Pipeline in Microsoft Fabric for Monthly Transactional Data

Srisakthi — Sun, 03 Nov 2024 06:29:24 GMT

Hi HamidBee ,

If you are more familiar with SQL you can go with Stored procedure activity. But for mre complex transformation its better to use Spark.

Stored Procedures vs. Notebooks for Transformations:

Typically for the complex transformation it is better to use Notebook(pyspark), moreover we have the flexibility to configure spark settings for the workload like setting your executors, dynamic scaling etc. for better performance and cost effective.

Monthly File Handling:

In data pipeline we have Triggers(Preview feature so check with your internal team), you can set the triggers based on file arrival in azure blob storage and have your file name in such a way for dynamic pipeline parameters. In case if you dont want to use Triggers then you can schedule your pipelne based on your monthly schedule and have parametized file name to pick up.

Performance Optimization:

Analyse the pattern, load and adjust spark configuration accordingly for cost effective and performance using Fabric metrics app.

Thanks,

Srisakthi

Re: Advice Needed: Building a Data Pipeline in Microsoft Fabric for Monthly Transactional Data

HamidBee — Mon, 04 Nov 2024 15:34:28 GMT

Thanks for sharing this information.