Apache Iceberg Go 0.7.0 Release

The Apache Iceberg community is pleased to announce version 0.7.0 of iceberg-go.

This release covers approximately four months of development since the 0.6.0 release in May 2026 and is the result of merging nearly 750 PRs from 53 contributors, including 33 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 close the gap with the Iceberg V3 table specification:

Geometry and geography types: geometry and geography types with schema plumbing, WKB encoding and GeoArrow conversion, PROJJSON CRS read/write support, geo bounds emitted in manifest data files and aggregated at the manifest level, and bounding-box predicate evaluation for geo pruning
Variant shredding: A shredded variant reader and an opt-in shredded variant writer with type-uniformity inference, bounds for shredded fields, and predicate pushdown over shredded fields using a new extract expression
Deletion vectors: The scanner deletion-vector read path, DV writes wired into the position-delete producer, DVs for partitioned V3 tables, merging DVs on repeated merge-on-read deletes, DV planning validations matching Java's DeleteFileIndex, and cross-client DV format compliance tests
Row lineage: Row lineage preserved through copy-on-write rewrites, first_row_id inherited for V1/V2-era manifests in V3 tables, and row-group pruning re-enabled with correct lineage tracking
Unknown transforms: Unknown partition and sort transforms can now be read for forward compatibility
Nanosecond timestamps: Nanosecond support across predicates, year/month and bucket transforms, Substrait literals, and partition values

Metadata Tables and Incremental Scans🔗

Table.Inspect gained a full set of metadata tables, and two new incremental scan types were added:

Metadata tables: history, snapshots, metadata_log_entries, refs, manifests, data_files, delete_files, entries, partitions, position_deletes, files, all_files, all_data_files, and all_delete_files, all_manifests, and all_entries
Incremental append scan: Table.NewIncrementalAppendScan plans data files added between two snapshots
Incremental changelog scan: Table.NewIncrementalChangelogScan produces file-level insert and delete tasks, along with delete-file-backed changelog task types as groundwork for row-level changelogs

REST Server-Side Scan Planning🔗

iceberg-go can now delegate scan planning to a REST catalog that supports the scan-planning endpoints. Local planning remains the default; remote planning is opt-in:

Foundations: Iceberg expression JSON parsing and the public scan-planning API surface
Client: Scan-planning client methods, a WaitForPlan poller, and decoding of REST scan tasks into FileScanTasks
Delegation: Complete server-side scan delegation, concurrent fetching of remote plan tasks, and retries, telemetry, and documentation
Correctness: A Java scan-planning integration suite and verification that REST planning produces tasks equivalent to local planning

Metrics Reporting🔗

A new metrics package adds opt-in reporting for scans and commits:

Reporter framework: A Reporter contract with built-in reporters, report and wire types, and reporter selection through catalog, table, and scan
Instrumentation: ScanReport emitted from scan planning and CommitReport emitted from commits, with environment context in commit reports
Reporters: A REST metrics reporter and an OpenTelemetry reporter, documented in a metrics reporting guide

Encryption Foundations🔗

Groundwork for Iceberg table encryption was added:

Interfaces: EncryptionManager and KeyManagementClient interfaces with an in-memory KMS
KMS registry: A KMS catalog-property registry selected through encryption.kms-type / encryption.kms-impl

IO Improvements🔗

Per-cloud packages: io/gocloud was split into per-cloud backend packages so applications only link the cloud SDKs they use
Azure: ManagedIdentityCredential support
S3: Connect timeout property, compatibility with GCS HMAC endpoints, and explicit s3.* credentials overriding the context AWS config
GCS: The GCS client authenticates with supplied credentials
Glue: aws-profile support for the Glue catalog in the CLI config

Performance🔗

More than 120 performance PRs landed in this release, touching nearly every part of the read, write, and planning paths. Some of the highlights:

Parallel reads: Large Parquet files are split into scan tasks by row-group byte range, controlled by read.split.target-size
Scan planning: Snapshot manifests cached across scans, compiled filter and pruning plans cached, scan columns projected during manifest reads, and row limits honored during local planning
Delete handling: Equality deletes indexed by partition and sequence, positional deletes indexed by path and partition, equality and position deletes loaded lazily per scan task, and equality deletes pruned by data-file metrics
Row-group pruning: Parquet row groups pruned by dictionary
Maintenance: Shared manifest reads deduplicated during snapshot expiration and orphan cleanup
Writes: Parquet dictionary encoding enabled by default, matching Java Iceberg

Arrow Go v18.8🔗

iceberg-go now depends on Apache Arrow Go v18.8.0, up from v18.6.0 in 0.6.0 (#1504, #2002), so applications importing both modules will resolve arrow-go to at least v18.8.0. The new compute.VariantGet kernel backs variant extract, with a zero-copy fast path for fields shredded to the queried type. Variant columns are now written with the canonical arrow.parquet.variant Arrow extension name; 0.7.0 reads both the canonical and legacy parquet.variant names, but older iceberg-go readers will not recognize the new name during a mixed-version rolling upgrade.

Bug Fixes🔗

Nearly 400 bug fixes landed in this release, many of them hardening validation, JSON decoding, and defensive copying across the public API. Notable fixes include:

Breaking Changes🔗

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

Scan concurrency option (#1462): the misspelled WitMaxConcurrency scan option was renamed to WithScanMaxConcurrency.
Transform interface (#1252): Transform gained a type-aware ToHumanStrType method, which external Transform implementations must add. ToHumanStr(any) is deprecated.
Hadoop catalog IO interfaces (#1185): the MkdirIO and ReadDirIO interfaces were removed in favor of MkdirAll and WalkDir.
Hadoop catalog cloud schemes (#1961): catalog/hadoop no longer registers cloud schemes as a side effect. Users on cloud storage must blank-import the matching io backend package. The existing io/gocloud symbols remain available as deprecated aliases.
Orphan cleanup (#1286): orphan file cleanup now requires the table's FileIO to implement ListableIO.
Parquet dictionary encoding (#931): data files are now written with dictionary encoding enabled by default.

New Contributors🔗

Welcome to all 33 first-time contributors: @mithilgirish, @paveon, @sungwy, @Revanth14, @rraulinio, @artbcf, @gerinsp, @jeremybarner, @strophy, @neu-hsc, @ebyhr, @abnobdoss, @AhmedDlshad007, @nanookclaw, @prafulla-kiran, @rasta-rocket, @the-onewho-knocks, @iremcaginyurtturk, @hankivstmb, @jx2lee, @1991santhu, @mattfaltyn, @dgvj-work, @dimakuz, @ryanworl, @technicolorbeat, @codeAnqiang-ma, @deepak-2605, @smaheshwar-pltr, @ajiteshsingh, @sgg, @KranzL, @shoemoney

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:

Try iceberg-go with your workloads and report any issues you encounter
Review the contributor guide
Look for good first issues

For more information, visit the iceberg-go repository.