ETL vs ELT: Why the T has shifted
summary
Argues that the industry moves from ETL to ELT because cloud warehouses like Snowflake BigQuery Redshift and Delta Lake separate storage and compute so teams load raw source data and run SQL transforms in-place rather than pre-transforming outside. Describes a working system: Fivetran provides pre-built connectors and pre-defined ERDs that replicate sources into the warehouse with incremental micro-batches and deduplication, an idempotent retryable pipeline that eliminates most custom ETL scripts and reduces engineering toil.
AI-written summary · about
chapters (12)
- 0:00Intro
- 0:30The Problem: Centralizing data
- 1:17Traditional Underpinnings
- 3:32The analytics challenge
- 4:36So what happens when
- 5:59Drawbacks of Legacy ETL
- 7:40A simple, modern data stack
- 8:04Step 01: Replication
- 8:44Outdated ETL vs Modern ELT
- 9:45Benefits of ELT
- 11:57Common Use Cases
- 13:10The Solution
related talks
38mReigniting the ETL vs. ELT Debate - Which Better Supports the Business?
01h 28mA Guide to the Modern Data Stack
29mETL and ELT, What's The Difference?
15mNext Gen ETL Generators
41mModern Data Architectures – What They Are and How to Implement?
39mLeaving ETLs Behind: Moving from a ‘Code Less’ Solution to a ‘Code Full’ Approach
55mThe ELT pattern in action with DBT and SQLMesh
39mWhat Data Engineering Can Learn from Front-end Engineering