Manage data transformations like software using only SQL
Data analysts write select statements, and dbt turns them into tables and views in the data warehouse. This allows software engineering practices to be applied directly to the data transformation process. This repository contains the source code for dbt v2.0, which has been completely rewritten in Rust.
Core changes in v2.0
dbt v2.0 is designed to significantly reduce parsing, compilation, and execution times in large projects. It applies a strict language specification that enforces accuracy at parse time to reduce errors. The installation process is simplified, distributing as a single binary without requiring a Python runtime or dependency management. v2.0 generates artifacts in Parquet format, making dbt projects easier to query and analyze. Existing JSON artifacts are still generated for backward compatibility.
Model-based data transformation
Analysts transform data by writing select statements, known as 'models'. dbt manages and visualizes how these models connect to one another. It clarifies relationships between models and ensures transformation quality through testing. The local documentation experience is also improved, allowing dbt docs to work smoothly even in large projects.
Supported environments and installation
dbt v2.0 and its drivers are compiled for specific operating systems and architectures. macOS and Linux support both x86-64 and ARM, while Windows supports only x86-64 and does not yet support ARM. Installation does not require Python environment configuration, allowing it to be used independently of existing data engineering stacks.
License and pre-use considerations
The source code in this repository is released under the Apache 2.0 license. The dbt product itself is a separate distribution released under the dbt Product License. While the main benefit of moving from v1 to v2.0 is improved parsing and compilation speed, compatibility with existing Python-based workflows must be verified. Windows ARM users are currently not supported, so they should consider v1 or a different architecture in that environment.