Alex Merced • 7/13/2026

Iceberg v3 Deletion Vectors and Faster DML

This article discusses the challenges of performing DML operations (updates, deletes) on immutable data lake file formats like Parquet, focusing on write amplification. It introduces Apache Iceberg v3's deletion vectors and merge-on-read approach as solutions, explaining the trade-off between faster writes and read-time merge costs. The article covers copy-on-write vs. merge-on-read, how Iceberg tracks row-level deletes, compaction patterns for high-churn tables, and implications for engines like Dremio. It is a technical deep dive into data lakehouse architecture and performance optimization.

0 comments

#Copy On Write #Data Lakes #Apache Iceberg