DuckDB ships DuckLake 1.0

A data lake is a pile of files in object storage; a lakehouse adds database rules on top so analysts can update, query, and track those files like tables. DuckDB said on April 13 that DuckLake v1.0 is now production-ready. (ducklake.select) DuckLake stores table metadata in a SQL database and stores the actual data in Parquet files. The DuckDB team said the v1.0 reference implementation ships in DuckDB v1.5.2, released the same day, with support for SQLite, PostgreSQL, and DuckDB catalogs. (ducklake.select) (duckdb.org) The basic pitch is that DuckLake uses a regular database as the catalog instead of keeping metadata in scattered files on object storage. DuckDB compared that setup to Delta Lake with Unity Catalog and Apache Iceberg with Lakekeeper, but said its format is “built on SQL” and comes with backward-compatibility guarantees at v1.0. (ducklake.select) That design targets a problem DuckDB users have run into as the project moved from single-machine analysis toward shared analytics systems. DuckDB’s own release note for DuckLake 1.0 described it as a “SQL-only lakehouse format” that enables “multiplayer DuckDB,” meaning multiple DuckDB instances can coordinate through one central catalog. (github.com) (ducklake.select) DuckLake has been moving toward this release for nearly a year. The team said it first published the DuckLake specification in May 2025, then added features including attaching existing Parquet files without a deep copy, support for geometry and variant data types, and Iceberg compatibility in version 0.3 on September 17, 2025. (ducklake.select 1) (ducklake.select 2) The reference extension’s public repository still describes the extension line as “experimental” because it reflects the older 0.x series, but the April 13 announcement says the specification itself is now stable and production-ready. The team also published a release calendar showing DuckLake specification releases are currently tied to ducklake extension releases. (github.com) (ducklake.select) The bigger argument behind DuckLake is not just speed but architecture. MotherDuck co-founder Jordan Tigani, writing in May 2025, said major cloud warehouses separate data in object storage from metadata in transactional databases, and argued lakehouse formats built entirely around object storage still miss some database semantics. (motherduck.com) DuckDB is betting that argument now has a stable format behind it. With DuckLake 1.0 bundled into DuckDB 1.5.2, the project is no longer presenting the idea as a sketch for local analytics, but as a format for long-lived shared data systems. (ducklake.select) (duckdb.org)