Where transformation happens.
WHAT ETL IS
Extract, transform, then load the transformed result.
WHAT ELT IS
Extract, load the raw data, then transform within the destination.
WHY ELT BECAME DOMINANT
Warehouses became cheap and powerful enough to do the transformation.
WHAT ELT PROVIDES
Raw data retained, so transformations can be corrected and rerun Transformation in SQL, accessible to more people No separate processing infrastructure
WHAT ETL STILL SUITS
Transformation required before storage, for privacy or size Sources too large to load raw Streaming transformation
WHAT THE PRACTICAL DEFAULT IS
Load raw, transform in place.
WHAT THAT MEANS FOR DESIGN
Ingestion becomes simple and generic; complexity moves to transformation.
WHY THAT IS AN IMPROVEMENT
Transformation is easier to test, version and correct than extraction.
WHAT TO KEEP IN INGESTION
Only what must happen before storage: masking of sensitive fields, and format conversion.
WHAT TO PUSH TO TRANSFORMATION
Business logic, joins, aggregation, derivation.
WHAT TO NEVER DO
Transform during extraction in a way that discards information.
WHY
You cannot recover what was never stored.
