.DUCKDB

Introduction

A .duckdb file is a database made by DuckDB, an analytical SQL database that, like SQLite, runs inside the program that uses it and keeps a whole database in one file. Where SQLite is built for many small reads and writes, DuckDB is built for queries over large tables, and stores its data by column.

DuckDB was first released in 2019 by researchers at CWI in Amsterdam. Its storage format changed between early releases; since version 1.0 in 2024 newer releases can read files written by older ones. Files are also given .db or .ddb.

Structure

The file is made of fixed-size blocks. It starts with a main header, whose bytes 8 to 11 are the text DUCK, followed by the storage version number. Two database headers come next; DuckDB writes them in turn, so one of them always points to the last complete checkpoint.

The rest of the file holds the catalogue of schemas and tables and the table data. Each table is split into row groups, and each column of a row group is stored and compressed on its own. While a database is open, recent changes are kept in a write-ahead log beside it, with .wal added to the file name.

Links