A developer has ported the Apache Cloudberry MPP engine to PostgreSQL 19 as an extension, rather than maintaining it as a separate fork. This approach allows users to run the latest version of PostgreSQL while utilizing Cloudberry's distributed query processing capabilities, including the ORCA optimizer and columnar storage.
The project required 12 days of development and 629 commits. The developer implemented 22 new extension points within the PostgreSQL kernel to enable distributed query planning and execution. The modified kernel remains binary-compatible with vanilla PostgreSQL 19, ensuring that standard extensions like PostGIS 3.7 and pgvector 0.8.8 function without modifications.
The extension maintains ACID compliance across the cluster and supports standard Greenplum features such as DISTRIBUTED BY, PARTITION BY, and Motion nodes. While it offers significant advantages for complex SQL queries on large datasets, it requires logical migration for existing Greenplum users rather than a simple pg_upgrade.
Performance benchmarks indicate that while the extension excels at complex joins and distributed transactions, it currently lags behind specialized analytical engines like ClickHouse or StarRocks in raw scan and aggregation speed due to PostgreSQL's non-vectorized executor. The project is currently an experimental effort.
