pg_mooncake

pg_mooncake is a Postgres extension for running real-time analytics directly on Postgres tables. It creates a columnstore mirror of a regular table in Apache Iceberg format, so analytical queries can run quickly against data that is only moments old. It is aimed at teams that want analytics on transactional data without building a separate warehouse pipeline.

Ingestion is handled by moonlink, which streams and batches inserts, updates and deletes into the mirror. Queries are accelerated by DuckDB, and you query the mirrored table like any other Postgres table. Because the data is stored in Iceberg, it can also be read by other query engines, which keeps the analytical copy open to the rest of a data stack.

The simplest way to try it is the Docker image, which starts Postgres with the extension installed. It can also be built from source with Rust, pgrx and DuckDB build tools for Postgres versions 14 through 18, and requires adding it to shared preload libraries and enabling logical replication. The project is written in Rust and licensed under MIT.