Apache Iceberg Rust 0.9.0 Release

The Apache Iceberg community is pleased to announce version 0.9.0 of iceberg-rust.

This release covers development work from early January through early March 2026 and is the result of merging 109 PRs from 28 contributors, including 8 new contributors. See the changelog for the complete list of changes.

iceberg-rust is a native Rust implementation of the Apache Iceberg specification, providing high-performance libraries for reading, writing, and managing Iceberg tables in Rust applications and through Python bindings (pyiceberg-core).

Release Highlights

Trait-Based Storage Architecture

The most significant change in this release is a new trait-based storage architecture that decouples iceberg-rust from any single storage backend. The new Storage trait provides a clean abstraction for file I/O, with integration into FileIO to make it the primary extension point for adding new storage backends.

This release includes two native storage implementations built on this trait:

• LocalFsStorage: A native local filesystem backend
• MemoryStorage: An in-memory backend useful for testing

The existing OpenDAL-based storage was moved to its own crate (iceberg-storage-opendal), keeping the core library lightweight while preserving full OpenDAL support as an optional dependency. See the Storage trait RFC for the design rationale.

Enhanced DataFusion Integration

DataFusion integration continues to expand with DDL and query optimization improvements:

• CREATE TABLE: Native support for creating Iceberg tables through DataFusion SQL, with schema validation for partition projection
• DROP TABLE: Support for dropping tables through DataFusion SQL
• LIMIT pushdown: Pushes row limits into Iceberg scans to avoid reading unnecessary data
• Predicate pushdown: Added support for Boolean, IsNaN, Timestamp, Binary, and LIKE/StartsWith predicates
• INSERT INTO: Support for inserting data into Iceberg tables through DataFusion SQL, with automatic sort-based clustering for partitioned writes

Reader Performance Improvements

The Arrow reader received several performance optimizations as part of an ongoing effort to reduce I/O overhead during reads:

• Byte range coalescing: Implemented get_byte_ranges on AsyncFileReader to coalesce small reads into fewer I/O operations
• Fast-path single-threaded reads: Optimized ArrowReader::read for single-concurrency scenarios by avoiding unnecessary async overhead
• Metadata size hints: Added Parquet metadata size hint option to ArrowReaderBuilder, eliminating unnecessary stat() calls when the metadata size is known
• File size propagation: Pass data file and delete file sizes to the reader to avoid redundant metadata lookups
• Reduced builder overhead: Avoided redundant create_parquet_record_batch_stream_builder() calls
• Timestamp support: Added timestamp type support in column creation

38-Digit Decimal Precision

The rust_decimal crate was replaced with fastnum, enabling full 38-digit decimal precision as required by the Iceberg specification. This also includes a fix for decimal byte representation used in hashing.

Bug Fixes

• Table upgrades: Fixed v2 to v3 table upgrades that previously failed
• Sort order validation: Reserved sort order IDs can no longer contain fields
• Partition field IDs: Fixed reuse of partition field IDs in AddSpec
• REST catalog security: Filtered sensitive headers from error logs to prevent credential leakage
• SqlCatalog: Fixed SqlCatalogBuilder not persisting the supplied catalog name
• S3Tables: Correctly interpret warehouse as table_location for S3Tables

Breaking Changes

This release includes several breaking changes:

• Storage architecture: The Storage trait is now integrated with FileIO, and OpenDAL storage moved to iceberg-storage-opendal
• MSRV: Bumped to Rust 1.92.0
• Decimal: Replaced rust_decimal with fastnum for 38-digit precision support

Dependency Updates

• Upgraded to DataFusion 52.2
• Various updates to Arrow, Parquet, and other dependencies

Infrastructure Improvements

• Security scanning: Added CodeQL workflow for GitHub Actions security analysis
• Workflow permissions: Added explicit least-privilege permissions to workflows
• Faster tests: Adopted nextest for faster test execution with parallel integration tests
• Python tooling: Switched to uv for Python dependency management

What's Next

Work is already underway on several features for upcoming releases:

• Storage trait expansion: With the new trait-based storage architecture landed in 0.9.0, the community is looking to integrate additional storage backends, such as object_store, making it easy for users to opt into alternative storage implementations.
• Table encryption: AES-GCM encryption support is in progress, bringing Iceberg's table-level encryption spec to Rust with Java-compatible ciphertext formats. See the table encryption RFC for the full design, including envelope encryption with a chained key hierarchy and KMS integration.

Getting Involved

The iceberg-rust project welcomes contributions. We use GitHub issues for tracking work and the Apache Iceberg Community Slack for discussions.

The easiest way to get started is to:

1. Try iceberg-rust with your workloads and report any issues you encounter
2. Review the contributor guide
3. Look for good first issues

For more information, visit the iceberg-rust repository.