Apache Iceberg v3: Deletion Vectors, VARIANT Type, and Upgrade Path
trending_up Trend: aws

Apache Iceberg v3: Deletion Vectors, VARIANT Type, and Upgrade Path

calendar_month June 26, 2026

Apache Iceberg v3 Reaches Production Readiness

With the release of Apache Iceberg 1.11.0 on May 19, 2026, the format version 3 (v3) of the open table format is officially production-ready. This major update introduces critical performance enhancements and new capabilities designed for modern enterprise lakehouse architectures, including binary deletion vectors, native VARIANT data types, default column values, and row-level lineage tracking.

Industry giants including Databricks, Dremio, Starburst, AWS, and Google Cloud are actively rolling out native support and upgrade paths for Iceberg v3.


Key Features in Apache Iceberg v3

  1. Binary Deletion Vectors: Iceberg v3 replaces the positional delete files used in v2 with compact, binary bitmaps per data file. In merge-on-read configurations, this dramatically lowers the read amplification during updates and deletes, accelerating CDC (Change Data Capture) pipelines.

  2. Native VARIANT Type: The new VARIANT type enables efficient storage of semi-structured JSON data without requiring schema flattening, combining the flexibility of document stores with the query performance of columnar formats.

  3. Geospatial Support (GEOMETRY/GEOGRAPHY): Native spatial data types allow organizations to run geospatial queries and joins directly within their data lakehouse.

  4. Default Column Values: Schema evolution is simplified by allowing default column values, eliminating the need for costly data backfills when new columns are added.

  5. Row-Level Lineage: Enhanced lineage tracking down to individual rows provides granular governance and auditability for compliance-heavy environments.


Upgrade Best Practices

While Iceberg v3 is designed to support seamless migrations, enterprise platforms recommend:

  • Incremental Migration: Since the format version is a table-level property, upgrade tables incrementally rather than all at once.
  • Verify Engine Readiness: Ensure your query engines (Spark 3.5+, Snowflake, Trino, Dremio) fully support reading and writing format version 3.
  • Pre-production Validation: Test the table migration on staging environments before upgrading production datasets.

Sources

  1. AWS announces support for Apache Iceberg V3 deletion vectors and row lineage
  2. Apache Iceberg v3: What Changed and How to Upgrade Safely
  3. Databricks AWS Official Docs
  4. Azure Databricks Official Docs
  5. Dremio Blog
  6. Starburst Blog
  7. Google Cloud Lakehouse Docs