What is a Data Warehouse?

Warehouses are built for a different workload than the operational databases that power an application: instead of many small reads and writes for a single record (an OLTP pattern), a warehouse is optimized for scanning and aggregating huge volumes of historical data across many rows at once (an OLAP pattern), which is why warehouse engines use column-oriented storage rather than the row-oriented storage typical of operational databases. Data gets into a warehouse through an ETL or ELT pipeline that extracts it from operational systems, cleans and conforms it to the warehouse's schema, and loads it - which is also why a warehouse's numbers can lag the live application by minutes or hours, a tradeoff acceptable for reporting but not for the operational system itself. A warehouse is usually the layer that sits above a data lake (which holds the raw, unstructured version) and below data marts (which hold department-specific curated slices), forming a pipeline from raw to structured to specialized.