Documents
Data Warehouse ETL Framework
Data Warehouse ETL Framework
  • Introduction To The ETL Framework
  • Framework Overview
    • Framework Description
    • Why You Need An ETL Framework
      • What is the ETL Framework?
      • Framework Objectives
      • Can I use my own resources to implement The Framework?
    • Download The Framework
    • What's In The Box
    • v1.0 Release
    • Training Resources For Engineers
    • Getting Help
  • General Principles
    • Never Finished
    • Overall Engineering Philosophy
    • Self Service Analytics As The Goal
    • Strict Adherence To Globally Accepted Practice
    • Data Flow Design
    • Effective DataOps
    • Code First Philosophy
    • Melding Kimball And Inmon
    • Master Data Management
    • Open Architecture
    • Decoupled Systems
    • Independent Identically Executing Processes
    • Robust Processes
    • Self-Diagnostics
    • Data Provenance
    • Adverse Reporting
    • Error Recovery
    • Archrival And Recovery
    • Documentation As A Necessity
  • DataOps For The Uninitiated
    • DataOps Initiation
    • What Is Agile Software Development?
    • What Is DevOps?
    • What Is Statistical Process Control?
    • You Take All That And You Wind Up With...
    • The DataOps Manifesto
    • Want To Learn More About DataOps?
  • ETL Developer's Field Guide
    • The Plain English Explanation Of What You Are About To Build
    • Data Warehouse User Profiles
    • Using Data Source Names
    • Why You Should Not Use SSIS
      • SSIS Is Not Forward Compatible
      • SSIS Is Not Very Performant
      • Python Development Is Faster Than SSIS
      • SSIS Is Difficult To Deploy To Production
      • SSIS Is Difficult To Maintain
    • Python Software Engineering Considerations
    • Enterprise Data Ecosystem
      • EDW's Place In Enterprise Architecture
      • EDW ETL Overview
    • ETL Environment Databases
    • Data Acquisition Paradigms
      • Demand Pull
      • Supply Push
    • The Common Model
    • The Semantic Layer
    • Database Object Naming Conventions
      • Columns
      • Indexes
      • Primary Keys
      • Schemas
      • Stored Procedures
      • Tables
      • User Defined Functions
    • ETL Reference Architecture
    • Table Anatomy
      • Sample Dimension Table
      • Sample Fact Table
      • Sample Indexed View
      • Sample Junk Dimension Table
      • Sample Master Data Management Table
      • Sample Stage Table
    • Master Data Management
      • MDM With Exact Match On Source System Key
      • Implementation Details
    • Slowly Changing Dimension Processing
      • General Process
      • Creating Empty Records
      • SCD Processing Types
      • Implementing SCDs As Temporal Tables
    • Handling Calculated Values
    • Warehouse Load Commandments
    • Code Style Guide
      • SSIS
      • Transact SQL
      • Python
      • C#
    • The Role Of Schemas
      • Standard Schemas And Their Definitions
      • Schemas As A Security Device
      • Schemas As A DB Object Differentiator
    • Feedback And Control Systems
      • Having An Engineering Mindset
      • The Mechanics Of Checking For Unprocessed Records
      • Time Series Analysis 101
      • Passive Monitoring System Design Theory
      • Passive Monitoring System Implementation
    • Source Control
    • Loading Historical Data
    • Security Access Model
    • Wrapping Up
  • ETL Environment Set Up
    • Install Anaconda
    • Create Initial Source Control Folders
    • Create Primary Databases And Schemas
    • Customize And Run Deployment Scripts
    • Install Python Packages
    • Create Data Source Name
    • Create The SSIS Catalogue
    • Create The File I/O Directory Structure
    • Create Global Environment In The Integration Services Catalog
    • Create The SQL Server Aliases
    • Create The BIAnalytics Proxy Account
    • Create The Opt Folder
    • Configure Email Alerts
  • Standard SSIS ETL Development Package
    • Starting A New Process
    • Tracking Package Variables
    • Remove Things You Do Not Need
    • Creating New Configuration Settings
  • Sample Script Guide
    • SQL
      • Finalize And Audit Scripts
        • 18 CREATE PROC usp_RecordRowCounts
        • 19 CREATE PROC usp_MarkRecordsAsProcessed
        • 20 CREATE PROC usp_CheckForUnprocessedRecords
      • Monitoring Scripts
        • 21 CREATE PROC usp_DisplayTablesNotLoading
        • 22 CREATE PROC usp_LoadTableLoadReportingTable
        • 23 CREATE PROC usp_TableLoadMonitoring
      • Table Object Sample Scripts
        • sample type I dimension table
        • sample type II dimension table
        • sample fact table
        • sample indexed view
        • sample junk dimension table
        • sample master data management table
        • sample stage table
      • Data Processing Sample Scripts
        • Data Pull
        • batch processing
        • bulk insert
        • data cleansing
        • fact table load
        • remove dups
        • type I dimension processing
        • type II dimension processing
      • Helper Scripts
        • Create Database
        • add rowhash
        • change collation
        • configuration insert sample script
        • documentation block
        • populate fact table with fake data
        • proc execution scripts
        • show all columns in database
        • troubleshooting
        • util
        • start an agent job from T-SQL
    • Python
      • Building Blocks
        • CombineCSVsIntoOneFile
        • ConvertCommaToPipeDelimitedFile
        • ConvertExcelToCSV
        • LoopingOverFilesInADirectory
        • Process Zip File
        • QueryDatabaseAndWriteSmallFile
        • QueryDatabaseAndWriteLargeFile
        • SendEmail
        • StringMatching
        • YAMLConfigImport
      • Full Solutions
        • DownloadMoveAndStoreDataCSV
        • LoadLargeCSVsIntoDataWarehouseStagingTables
        • MoveAndStoreDataExcel
        • ReloadFromArchive
      • Jupyter Notebooks
        • Passive Monitoring System Design Theory
    • PySpark
      • ConnectToSpark
      • LoadCSV
      • ExampleDataProcessing
    • Windows Batch
  • A Methodology To Rapidly Convert OLTP Databases to OLAP Solutions
    • Step 1: Find The Nouns
    • Step 2: Find The Stuff We Want To Do Math On
    • Step 3: Analyze Relationships
  • Data Model Creation Tool
    • Record Count
    • Sample Data
    • Create Stage Table
    • Char Length Analysis
    • Column Notes
    • Column Cleansing Notes
    • Source To Target Mapping
    • Dimension List
    • Fact Table Creation Helper
    • Foreign Key Creation
    • Process Fact Script Helper
    • View Creation Helper
    • Date Role Play View Helper
    • Data Model View Creation Helper
    • List Population Values
  • Performance Monitoring
    • The Daily Chore
    • Diagnostic Tools
      • Using The Built-In Stored Procs
      • Using The Built-In Views
    • Logging Database Diagram
  • Data Warehouse Troubleshooting Guide
    • Generalized Troubleshooting Steps
  • Business Analytics Capability Maturity Model
    • What Is A Business Analytics Capability Maturity Model?
    • Level 0. Operational Reporting
    • Level 1. Rapid Delivery
    • Level 2. Self Service
    • Level 3. Central Repository
    • Level 4. Open Data 1
    • Level 5. Open Data 2
    • Level 6. Feedback 1
    • Level 7. Data Archaeology
    • Level 8. Crystal Ball
    • Level 9. Feedback 2
    • Level 10. Oil Rig
    • Level 11. The Singularity
  • Appendices
    • Appendix A. What is Medium Data®?
    • Appendix B. The Benefits Of Using Python And T-SQL Over SSIS For ETL
Powered by GitBook
On this page
  1. ETL Developer's Field Guide

Handling Calculated Values

PreviousImplementing SCDs As Temporal TablesNextWarehouse Load Commandments

Last updated 4 years ago

In general, there should be no math in the pipeline. You should not be storing the results of calculations. instead, you should be storing the components of calculations. The reason for this is because the definitions of values often change. This is particularly true of greenfield projects as the business negotiates with various internal factions and consensus is built around the definition of various KPIs. Even after this process is complete, it is a poor assumption that the definition of a KPI is written in stone.

Having math in your pipeline is problematic. The question is, where does the calculation go?

  1. Does it go in the pull?

  2. Does it go in the cleansing process?

  3. Does it go in the fact table load?

  4. Do you have a published policy about where calculations go?

  5. Are all your engineers familiar with the policy so they don't put calculations where they don't belong according to the policy?

  6. Are you 100% certain of 5 because now there is a problem with numbers and you can't find where the calculation lives.

Having math in the pipeline is bad DataOps. It is one of the very many opportunities you will be given to shoot your organization's analytic efforts in the foot. Anything that slows down your ability to push code needs to be tossed aside in favor of something simpler, faster, and easier to manage.

Having all calculations handled in BI Reporting tools or processes in the Reporting database is a natural thought process that does not require a policy for everybody to memorize. Because of course that's where the math goes. Where else would it go but the end point where the value can be surfaced at runtime just like we all learned in college but somehow forgot when we hit the real world. I guess all this money made us forget a few fundamentals.

Having said all that, this is software engineering and edge cases are a thing. There are very good arguments to be had for having math in the pipeline if it makes sense to do so. The scenario that I run into most often is that the result of a calculation is itself a component of another calculation and we do not care one bit about the original components. In that case, it makes sense to put math in the pull.

The tl;dr version:

  1. Put math in at the end points of your pipeline. Do not put them in the middle all helter skelter.

  2. When deciding where to put calculations, use the following heuristic.

    1. Is it a day ending in "y"? Put the math in the report.

    2. There is no second rule.