Lakehouse

A lakehouse is an architectural approach that combines lake-style storage, commonly in open file formats, with table management and analytical capabilities associated with warehouses. The 2021 CIDR paper describes this combination as a way to support both SQL analytics and data science on shared data.

What is added to a collection of files?

A table layer tracks which files belong to a table version and coordinates supported updates. With transactional commits, readers can see a committed table state instead of an arbitrary mixture of files from an unfinished write. Metadata, query optimization, and access controls help turn storage into an analytical system.

Consider a sales table being refreshed while a report reads it. A suitable implementation can let the report read the previous committed snapshot until the new one is ready. Merely placing files in object storage does not establish that behavior; the table format, engine, and write protocol must work together.

A combined architecture still needs engineering

A lakehouse aims to reduce the need for separate copies maintained only to serve different workloads. It does not prohibit copies, and it does not eliminate transformations. Refund definitions, duplicate handling, and quality checks still have to be designed.

Support for transactions, historical versions, and interoperability varies across implementations. The lakehouse label alone is not a guarantee of lower cost or better query performance. Compare actual workloads and operational requirements. A lake emphasizes flexible storage, a warehouse emphasizes managed analysis, and a lakehouse deliberately combines those roles.

Reference: Armbrust et al.: Lakehouse, CIDR 2021.