Why warehouse versus lake is over

Welcome back. I'm Param. This is the AWS lakehouse course in the learnwithparam data engineering track. We are going to build the real thing: Lambda extracts to S3, Glue writes Iceberg, Snowflake reads it without copying, Airflow runs the show, Terraform owns the infrastructure. Everything grounds in a real companion project: Lambda, Glue, Snowflake, Airflow, and Terraform code you will read and reason about piece by piece.

Analytics infra has gone through three eras. First, the warehouse: tightly-coupled storage and compute, fast queries, hard to share data. Then the lake: cheap S3 plus Spark, flexible but loose, schema drift everywhere. Now the lakehouse: one storage layer with table formats, multiple engines, the discipline of a warehouse with the openness of a lake.

Warehouse, lake, lakehouse

Lakehouse keeps storage open and lets multiple engines read it through a shared table format.

Snowflake is excellent at queries. It is expensive at storage. Storing all your raw data in Snowflake-managed tables doubles every byte: once in S3, once in Snowflake. The lakehouse keeps storage in S3 and lets Snowflake read it on demand. You get Snowflake compute and S3 storage costs, not both.

Quiz: Quiz

Loading practice…

AI prompt: Try it: classify your current stack

Loading practice…