Data warehouse vs data lake

WHAT THIS TESTS The interviewer wants a clear grasp of two complementary storage paradigms and when each fits, not a vague preference for one.

A GOOD ANSWER COVERS A data warehouse stores structured, cleaned, and modeled data, typically relational tables, and enforces schema-on-write: data must conform to a defined schema before it is loaded. It is optimized for fast, repeatable SQL analytics and business intelligence dashboards, and it tends to be more expensive per terabyte because storage and compute are tuned for query performance. A data lake stores raw data of any kind, structured rows, semi-structured JSON or logs, and unstructured images or video, on cheap object storage. It uses schema-on-read: no structure is enforced at ingest, and you impose schema only when you query. This makes lakes ideal for exploratory analysis, data science, and machine learning where future questions are unknown.

KEY DIFFERENCES Structure: warehouse is structured and curated, lake holds everything raw. Schema: warehouse is schema-on-write, lake is schema-on-read. Use cases: warehouse for BI and reporting, lake for ML and ad-hoc exploration. Cost and governance: lakes are cheaper but risk becoming ungoverned data swamps without cataloging and quality controls.

COMMON WRONG ANSWERS Saying a lake is simply a cheaper warehouse, ignoring schema timing and governance. Claiming warehouses cannot handle large scale, or that lakes always replace warehouses; many architectures use both, often as a lakehouse.

LIKELY FOLLOW-UPS What is a lakehouse. What is a data swamp and how do you avoid it. When would you use both together.

ONE CONCRETE EXAMPLE A retailer loads cleaned daily sales into a Redshift warehouse for finance dashboards, while dumping raw clickstream, app logs, and product images into an S3 data lake that data scientists query with schema-on-read to train a recommendation model.