Apache Iceberg V3

Iceberg V3 deletion vectors and row lineage 2024.

Definition

Iceberg V3 (TableSpec v3) adds: default values per column, geospatial WKB type, variant type (semi-structured), row lineage tracking, multi-arg transforms partition, Position and Equality deletion vectors (Puffin format). Backward compatible with Iceberg V2 (but V3 features require V3 readers/writers). Adopted by Snowflake, Databricks, AWS Athena, Google BigQuery (preview), Apache Spark, Trino.

Origin

Netflix 2017 then Apache Software Foundation Top-Level Project 2020. V1 2018, V2 2021, V3 spec 2024.

Usage

Emerging dominant data lakehouse format 2024+. Direct competitor of Delta Lake (Databricks) and Apache Hudi.

Related terms

• Apache Iceberg — fiche socle.
• Delta Lake V4 — concurrent direct.
• Apache Hudi — concurrent.