DuckDB v2.0 Preview: Server Mode and Stable C API Headline ‘Cyanoptera’ Release

The DuckDB Foundation has previewed version 2.0 of its in-process SQL OLAP database management system, codenamed “Cyanoptera” after the cinnamon teal. Scheduled for release this autumn, the update represents a major architectural shift for the open-source project, moving beyond its traditional single-user, in-process model to support native client/server deployments. The release follows more than 10,000 commits since version 1.5 and introduces a suite of technical changes designed to enhance performance, extensibility, and network connectivity.

Central to the v2.0 roadmap is the Quack extension, which graduates to stable status to implement DuckDB’s native protocol for client/server interactions. This allows any DuckDB process to serve databases over a network, with other instances attaching via the new CONNECT statement. The update also introduces a remote pushdown optimizer capable of routing SQL directly to PostgreSQL and MySQL databases, enabling efficient query execution across heterogeneous environments. This shift positions DuckDB to compete more effectively in multi-tenant, long-running deployments that require transactional isolation and full multi-connection support.

Performance improvements are driven by the introduction of asynchronous I/O throughout the engine, allowing input and output operations to scale independently of query processing. This architecture significantly boosts parallelism for remote reads, particularly on network storage systems such as Amazon S3. The default storage format has been updated to version 2.0.0, featuring buffer-managed ART indexes that are no longer pinned in memory. This change enables large indexed tables to open instantly with indexes paged in on demand, while lazy loading of column metadata and default DICT_FSST string compression further reduce memory overhead.

The release also marks a significant overhaul of the software’s foundational components, including the replacement of the PostgreSQL-derived SQL parser with a new PEG-based parser. This modernisation allows extensions to hook directly into the grammar, facilitating the development of new SQL syntax. Additionally, the C API has been stabilised and generated from a declarative YAML specification, ensuring that extensions built against it remain compatible across future versions without requiring re-targeting. The foundation has also introduced a repository registration system, allowing organisations to host and sign their own extensions using RSA public keys.

Further technical enhancements include the introduction of triggers, support for the VARIANT type to handle semi-structured data more efficiently, and NEAREST joins for vector similarity search. The ICU library has been removed from the core distribution, with timezone and collation support now handled by a separate, lighter icu extension. To guide the project’s future direction, the DuckDB Foundation will establish a stakeholder advisory board starting this autumn, providing key stakeholders with input on the development roadmap for DuckDB, DuckLake, and Quack.