Implementing Positional Deletes in Iceberg v3: Streamlining Merge-on-Read for Fast-Inbound Event Lakes

This article dives into Apache Iceberg v3's positional deletes and merge-on-read strategies for managing row-level changes in event lakes. It contrasts copy-on-write designs, which rewrite entire files for corrections, with Iceberg's delete-file support that enables efficient updates without full rewrites. The piece covers practical use cases like late-arriving events, privacy removals, and retractions, along with query planning, compaction, and engine compatibility considerations. It emphasizes that merge-on-read is not a universal speedup but a strategic choice for high-ingestion workloads, offering a balanced evaluation checklist for production deployment.