Apache Iceberg Go 0.6.0 Release

The Apache Iceberg community is pleased to announce version 0.6.0 of iceberg-go.

This release covers approximately three months of development since the 0.5.0 release in March 2026 and is the result of merging nearly 200 PRs from 40 contributors, including 26 first-time contributors. See the full changelog for the complete list of changes.

iceberg-go is a native Go implementation of the Apache Iceberg table format, providing libraries for reading, writing, and managing Iceberg tables in Go applications.

Release Highlights

Iceberg V3 Table Spec Support

This release continues to advance iceberg-go's implementation of the Iceberg V3 table specification:

• Variant type: Support for the V3 variant type (non-shredding) was added
• Deletion vectors: A deletion vector reader and SerializeDV/DVWriter authoring APIs, with blob header validation on read, cardinality validation through ReadDV, and required blob properties enforced on write
• Row lineage: Row lineage support in V3, with reserved row-lineage fields projected as null when a file lacks them
• Nanosecond timestamps: Parquet support for the nanosecond timestamp types and mapping of nanosecond Arrow timestamps to the V3 TimestampNs/TimestampTzNs types
• Defaults: write-default and initial-default implemented for V3 and enforced for required fields
• Format-version gating: V3-only metadata fields are gated on sub-V3 reads, V2-only fields are gated on V1 reads, and the remaining V3 metadata-upgrade validation rules were ported
• Format-version upgrade: A Transaction.UpgradeFormatVersion API was added

Row-Level Deletes

iceberg-go gained a full equality-delete path and an atomic row-level mutation API:

• Equality deletes: Write path, read path in the scanner, and support for partitioned tables
• RowDelta API: An atomic row-level mutation API was added
• Overwrite with deletes: Delete file removal in overwrite commits and deletion vectors classified in scan planning

Table Maintenance and Compaction

A new compaction and maintenance suite landed in this release:

• Compaction execution: RewriteDataFiles for compaction with a bin-pack strategy and Analyze for dry-run planning
• Dangling deletes: Dangling equality deletes are dropped during RewriteDataFiles
• RewriteFiles op: A RewriteFiles snapshot-op builder was added

Concurrency and Conflict Resolution

Optimistic concurrency control was substantially hardened:

• Conflict validation framework: A conflict validation framework was added and wired into producers
• Commit retries: doCommit retries on commit-conflict errors, with refresh-and-replay between retries and concurrent-write conflicts wrapped with ErrCommitFailed
• RowDelta conflicts: RowDelta uses a partition-scoped conflict check instead of always-true

Hadoop Catalog

A Hadoop catalog implementation was added over the course of the release:

• Scaffold and path helpers: Initial Hadoop catalog scaffold
• Namespace and table operations: Namespace operations, list/drop/rename, and table and namespace CRUD
• CLI integration: Hadoop catalog CLI integration

Catalog Improvements

• Multi-table commits: A TransactionalCatalog interface and REST multi-table commit, plus CommitAndReload
• OAuth and credentials: OAuth token refresh, vended credential refresh, audience and resource OAuth params, and a WithOAuthTLSConfig option for separate OAuth-server TLS
• Hive catalog: Register table and create view support
• REST views: RegisterView was added to the REST catalog

Expanded CLI

The command-line tool gained a large set of new commands:

• Maintenance commands: expire-snapshots, clean-orphan-files, partition-stats, and compact analyze/compact run
• Table operations: branch create/tag create, upgrade/rollback, snapshots/refs, and an info summary command
• Schema flags: --schema-from-file to infer a schema from Parquet files and a --show-defaults flag
• Argument parsing: The CLI migrated from docopt-go to go-arg

Write Performance and Bloom Filters

• File sizing: Parquet files are rolled based on actual compressed size
• Partitioned writes: A clustered partitioned write path was added
• Memory pressure: Reduced memory pressure by releasing records and an optional skip of the duplicate check on AddDataFiles
• Bloom filters: Bloom filter scan pruning for Parquet row groups and bloom filter properties wired into the Parquet writer
• Metadata compression: Support for the zstd metadata compression codec

IO Improvements

• In-memory FS: A native in-memory filesystem was added
• Aliyun OSS: The oss scheme is registered for Aliyun OSS S3-compatible storage
• New interfaces: A ListableIO interface replaces reflect-based directory walking, and a BulkRemovableIO interface was added
• S3 Table Buckets: PermanentRedirect is resolved for S3 Table Buckets

Statistics

• Statistics updates: SetStatistics, RemoveStatistics, AddEncryptionKey, and RemoveEncryptionKey updates were implemented
• Partition statistics: SetPartitionStatisticsUpdate and RemovePartitionStatisticsUpdate were added

Bug Fixes

Notable bug fixes in this release include:

• Fix a nil field Type panic in MarshalJSON
• Use floored division in HourTransform for pre-epoch timestamps
• Handle LargeString columns in partition transforms
• Fast-append must inherit all parent manifests unconditionally
• Fix a goroutine leak in positionDeleteRecordsToDataFiles
• Fix a deadlock in MapExec when workers error
• Fix a refcount leak in enrichRecordsWithPosDeleteFields
• Reject duplicate field IDs at schema construction
• Change equality_ids Avro element type from Long to Integer to match the spec
• Decode pre-1.4 Java Iceberg legacy manifest list field names
• Resolve a data race in MarshalJSON on a shared *Schema
• Emit a proper Avro fixed schema for FixedType partition columns

Breaking Changes

Some of these changes are breaking changes that need to be called out:

• Manifest entries iterator (#985): FetchEntries was replaced with an iterator-based approach.
• Multi-arg transforms (#824): partition and sort fields now support multi-argument transforms, changing the relevant field signatures.

New Contributors

Welcome to all 26 first-time contributors: @starpact, @PranjalChaitanya, @rockwotj, @RSP22, @Hashcode-Ankit, @hcrosse, @Herrtian, @gabrnavarro, @Narwhal-fish, @alliasgher, @abhirathod95, @rohilsurana, @twmb, @cassio-paesleme, @Jeffail, @benbellick, @varun0630, @hectar-glitches, @swjtu-zhanglei, @jacobmarble, @C-Loftus, @nssalian, @tanmayrauth, @fallintoplace, @happydave1, @mzzz-zzm

Getting Involved

The iceberg-go 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-go 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-go repository.