When Paimon Beats Iceberg for Mutable Streams

This article analyzes when Apache Paimon outperforms Apache Iceberg for mutable data streams, particularly high-churn CDC (Change Data Capture) workloads. It explains Paimon's Log-Structured Merge-tree (LSM-tree) architecture, designed for continuous upserts and Flink-native execution, versus Iceberg's merge-on-read approach that accumulates delete files. The article covers Paimon's primary key tables, changelog streams, and practical configuration for high-churn scenarios, while also noting where Iceberg remains superior for append-heavy analytics. It provides a workload decision matrix to help choose the right format.