A Preview of DuckDB v2.0

DuckDB v2.0, codenamed "Cyanoptera," is set to be released this fall. This major version brings a host of exciting new features. For starters, DuckDB will now operate as a server, enabling it to serve databases over the network via the QUACK extension. This server mode is achieved through a new SQL statement called CONNECT, which allows any DuckDB process to attach to and route queries to a remote database.

One of the most significant changes is a new SQL parser and a reworked C API. These changes include carefully chosen breaking modifications that necessitate the major version bump. In terms of storage, DuckDB will now use a new default storage format.

There's also a new data type called VARIANT, which functions like JSON on steroids. VARIANT can store differently-shaped data in every row, unlike JSON, and it automatically detects common structure in semi-structured data to compress well in storage and execute queries quickly.

DuckDB v2.0 also introduces triggers, a feature requested for a long time. These include BEFORE and AFTER triggers, FOR EACH ROW and FOR EACH STATEMENT, transition tables, multiple triggers per event, RETURNING on triggered tables, and DROP TRIGGER. Triggers are designed to work well with long-running DuckDB services.

The new SQL dialect of DuckDB gets several additions as well, such as NEAREST joins, DML inside CTEs, nested schemas, variable syntax, JSON mutation functions, and recursive CTEs with USING KEY aggregation.