DuckDB is a high-performance, embeddable analytical database system that has gained massive popularity in the last few years. It allows you to accomplish a surprising number of analytical tasks blazingly fast since it has a state-of-the-art vectorized query engine and it utilizes the local compute on your laptop to run queries. It is frequently referred to by data professionals as the “SQLite for analytics” for its simplicity and an embeddable model.

For developers, DuckDB provides programmatic access to its entire codebase through its extension model. It supports various data scanners (e.g. parquet, CSV, arrow, JSON) and over 500 scalar/aggregation functions through both in-tree and out-of-tree extensions. Any developer can write their own extension which supports specialized logic and/or functions for their needs. In this talk, we will take a look at how DuckDB extensions works and discuss some best practices and considerations around building DuckDB extensions from experience.

We will also see how we are able to extend the DuckDB extension as far as performing hybrid query execution – a query execution model that allows us to run queries closer to where the data lives in order to achieve better query performance, reduce cost, and give users more flexibility to decide where to run their queries. We will delve into the architecture of DuckDB’s query execution and how we are building out a delightful hybrid execution experience by both leveraging the existing DuckDB query execution flow and contributing to the DuckDB codebase. By the end of the talk, you will gain a better understanding of the power of DuckDB extensions, how query execution works and how to build an extension for your own needs.