DataFusion
DataFusion is a very fast, extensible query engine for building high-quality data-centric systems in
Rust, using the Apache Arrow
in-memory format.
DataFusion offers SQL and Dataframe APIs, excellent performance, built-in support for CSV, Parquet, JSON, and Avro, extensive customization, and a great community.
Features
- Feature-rich SQL support and DataFrame API
- Blazingly fast, vectorized, multi-threaded, streaming execution engine.
- Native support for Parquet, CSV, JSON, and Avro file formats. Support
for custom file formats and non file datasources via the
TableProvider trait.
- Many extension points: user defined scalar/aggregate/window functions, DataSources, SQL,
other query languages, custom plan and execution nodes, optimizer passes, and more.
- Streaming, asynchronous IO directly from popular object stores, including AWS S3,
Azure Blob Storage, and Google Cloud Storage. Other storage systems are supported via the
ObjectStore trait.
- Excellent Documentation and a
welcoming community.
- A state of the art query optimizer with projection and filter pushdown, sort aware optimizations,
automatic join reordering, expression coercion, and more.
- Permissive Apache 2.0 License, Apache Software Foundation governance
- Written in Rust, a modern system language with development
productivity similar to Java or Golang, the performance of C++, and
loved by programmers everywhere.
- Support for Substrait for query plan serialization, making it easier to integrate DataFusion
with other projects, and to pass plans across language boundaries.
Use Cases
DataFusion can be used without modification as an embedded SQL
engine or can be customized and used as a foundation for
building new systems. Here are some examples of systems built using DataFusion:
- Specialized Analytical Database systems such as [CeresDB] and more general Apache Spark like system such a [Ballista].
- New query language engines such as [prql-query] and accelerators such as [VegaFusion]
- Research platform for new Database Systems, such as [Flock]
- SQL support to another library, such as [dask sql]
- Streaming data platforms such as [Synnada]
- Tools for reading / sorting / transcoding Parquet, CSV, AVRO, and JSON files such as [qv]
- A faster Spark runtime replacement [Blaze]
By using DataFusion, the projects are freed to focus on their specific
features, and avoid reimplementing general (but still necessary)
features such as an expression representation, standard optimizations,
execution plans, file format support, etc.
Why DataFusion?
- High Performance: Leveraging Rust and Arrow's memory model, DataFusion is very fast.
- Easy to Connect: Being part of the Apache Arrow ecosystem (Arrow, Parquet and Flight), DataFusion works well with the rest of the big data ecosystem
- Easy to Embed: Allowing extension at almost any point in its design, DataFusion can be tailored for your specific use case
- High Quality: Extensively tested, both by itself and with the rest of the Arrow ecosystem, DataFusion can be used as the foundation for production systems.
Comparisons with other projects
When compared to similar systems, DataFusion typically is:
- Targeted at developers, rather than end users / data scientists.
- Designed to be embedded, rather than a complete file based SQL system.
- Governed by the Apache Software Foundation process, rather than a single company or individual.
- Implemented in
Rust, rather than C/C++
Here is a comparison with similar projects that may help understand
when DataFusion might be be suitable and unsuitable for your needs:
DuckDB is an open source, in process analytic database.
Like DataFusion, it supports very fast execution, both from its custom file format
and directly from parquet files. Unlike DataFusion, it is written in C/C++ and it
is primarily used directly by users as a serverless database and query system rather
than as a library for building such database systems.
Polars: Polars is one of the fastest DataFrame
libraries at the time of writing. Like DataFusion, it is also
written in Rust and uses the Apache Arrow memory model, but unlike
DataFusion it does not provide SQL nor as many extension points.
Facebook Velox
is an execution engine. Like DataFusion, Velox aims to
provide a reusable foundation for building database-like systems. Unlike DataFusion,
it is written in C/C++ and does not include a SQL frontend or planning /optimization
framework.
Databend is a complete
database system. Like DataFusion it is also written in Rust and
utilizes the Apache Arrow memory model, but unlike DataFusion it
targets end-users rather than developers of other database systems.
DataFusion Community Extensions
There are a number of community projects that extend DataFusion or
provide integrations with other systems.
Language Bindings
Integrations
Known Uses
Here are some of the projects known to use DataFusion:
Examples
Please see the example usage in the user guide and the datafusion-examples crate for more information on how to use DataFusion.
Roadmap
Please see Roadmap for information of where the project is headed.
Architecture Overview
There is no formal document describing DataFusion's architecture yet, but the following presentations offer a good overview of its different components and how they interact together.
- (July 2022): DataFusion and Arrow: Supercharge Your Data Analytical Tool with a Rusty Query Engine: recording and slides
- (March 2021): The DataFusion architecture is described in Query Engine Design and the Rust-Based DataFusion in Apache Arrow: recording (DataFusion content starts ~ 15 minutes in) and slides
- (February 2021): How DataFusion is used within the Ballista Project is described in *Ballista: Distributed Compute with Rust and Apache Arrow: recording
User Guide
Please see User Guide for more information about DataFusion.
Contributor Guide
Please see Contributor Guide for information about contributing to DataFusion.