Apache Arrow Rust 57.0.0 Release

Published 30 Oct 2025
By The Apache Arrow PMC (pmc)

The Apache Arrow team is pleased to announce that the v57.0.0 release of Apache Arrow Rust is now available on crates.io (arrow and parquet) and as source download.

See the 57.0.0 changelog for a full list of changes.

New Features

Note: Arrow Rust hosts the development of the parquet crate, a high performance Rust implementation of Apache Parquet.

Performance: 4x Faster Parquet Metadata Parsing 🚀

Ed Seidl (@etseidl) and Jörn Horstmann (@jhorstmann) contributed a rewritten thrift metadata parser for Parquet files which is almost 4x faster than the previous parser based on the thrift crate. This is especially exciting for low latency use cases and reading Parquet files with large amounts of metadata (e.g. many row groups or columns). See the blog post about the new Parquet metadata parser for more details.

Figure 1: Performance improvements of Apache Parquet metadata parsing between version 56.2.0 and 57.0.0.

New arrow-avro Crate

The 57.0.0 release introduces a new arrow-avro crate contributed by @jecsand838 and @nathaniel-d-ef that provides much more efficient conversion between Apache Avro and Arrow RecordBatches, as well as broader feature support.

Previously, Arrow‑based systems that read or wrote Avro data typically used the general‑purpose apache-avro crate. While mature and feature‑complete, its row-oriented API does not support features such as projection pushdown or vectorized execution. The new arrow-avro crate supports these features efficiently by converting Avro data directly into Arrow's columnar format.

See the blog post about adding arrow-avro for more details.

High-level `arrow-avro` architecture

Figure 2: Architecture of the arrow-avro crate.

Parquet Variant Support 🧬

The Apache Parquet project recently added a new Variant type for representing semi-structured data. The 57.0.0 release includes support for reading and writing both normal and shredded Variant values to and from Parquet files. It also includes parquet-variant, a complete library for working with Variant values, VariantArray for working with arrays of Variant values in Apache Arrow, computation kernels for converting to/from JSON and Arrow types, extracting paths, and shredding values.

 // Use the VariantArrayBuilder to build a VariantArray
let mut builder = VariantArrayBuilder::new(3);
builder.new_object().with_field("name", "Alice").finish(); // row 1: {"name": "Alice"}
builder.append_value("such wow"); // row 2: "such wow" (a string)
let array = builder.build();

// Since VariantArray is an ExtensionType, it needs to be converted
// to an ArrayRef and Field with the appropriate metadata
// before it can be written to a Parquet file
let field = array.field("data");
let array = ArrayRef::from(array);
// create a RecordBatch with the VariantArray
let schema = Schema::new(vec![field]);
let batch = RecordBatch::try_new(Arc::new(schema), vec![array])?;

// Now you can write the RecordBatch to the Parquet file, as normal
let file = std::fs::File::create("variant.parquet")?;
let mut writer = ArrowWriter::try_new(file, batch.schema(), None)?;
writer.write(&batch)?;
writer.close()?;

This support is being integrated into query engines, such as @friendlymatthew's datafusion-variant crate to integrate into DataFusion and delta-rs. While this support is still experimental, we believe the APIs are mostly complete and do not expect major changes. Please consider trying it out and providing feedback and improvements.

Thanks to the many contributors who made this possible, including:

See the ticket Variant type support in Parquet #6736 for more details

Parquet Geometry Support 🗺️

The 57.0.0 release also includes support for reading and writing Parquet Geometry types, GEOMETRY and GEOGRAPHY, including GeospatialStatistics contributed by Kyle Barron (@kylebarron), Dewey Dunnington (@paleolimbot), Kaushik Srinivasan (@kaushiksrini), and Blake Orth (@BlakeOrth).

Please see the Implement Geometry and Geography type support in Parquet tracking ticket for more details.

Thanks to Our Contributors

$ git shortlog -sn 56.0.0..57.0.0
    36  Matthijs Brobbel
    20  Andrew Lamb
    13  Ryan Johnson
    11  Ed Seidl
    10  Connor Sanders
     8  Alex Huang
     5  Emil Ernerfeldt
     5  Liam Bao
     5  Matthew Kim
     4  nathaniel-d-ef
     3  Raz Luvaton
     3  albertlockett
     3  dependabot[bot]
     3  mwish
     2  Ben Ye
     2  Congxian Qiu
     2  Dewey Dunnington
     2  Kyle Barron
     2  Lilian Maurel
     2  Mark Nash
     2  Nuno Faria
     2  Pepijn Van Eeckhoudt
     2  Tobias Schwarzinger
     2  lichuang
     1  Adam Gutglick
     1  Adam Reeve
     1  Alex Stephen
     1  Chen Chongchen
     1  Jack
     1  Jeffrey Vo
     1  Jörn Horstmann
     1  Kaushik Srinivasan
     1  Li Jiaying
     1  Lin Yihai
     1  Marco Neumann
     1  Piotr Findeisen
     1  Piotr Srebrny
     1  Samuele Resca
     1  Van De Bio
     1  Yan Tingwang
     1  ding-young
     1  kosiew
     1  张林伟