• Building a DuckDB Alternative in Postgres: Introduces pg_lakehouse, a new extension that turns Postgres into a DuckDB alternative. It allows direct querying of external object stores like S3, with table formats like Delta Lake and file formats like Parquet. Goals are fast analytics over data lakes with only Postgres and easy data joining/moving between Postgres and data lakes.
• Putting a Foreign Data Wrapper on Steroids: Uses the foreign data wrapper API in Postgres to connect to external data sources. Differentiates by being highly performant (queries pushed to Apache DataFusion, 8X faster than Elasticsearch) and having broad coverage (uses Apache OpenDAL for 40+ data stores, unlike specific wrappers for one format/store).
• How It's Built: Uses two Postgres APIs - the executor hook and foreign data wrapper. The executor hook reroutes foreign table queries to DataFusion. If a query fails in DataFusion, it falls back to Postgres. Users can determine the query engine used with EXPLAIN.
• Getting Started: Provides an example Parquet file in a public S3 bucket. After installing the extension, code is shown to create a foreign data wrapper, server, and foreign table to query the dataset. Instructions for connecting own object stores are in the documentation.
• What's Next: Focuses on write support to centralize data lake operations in Postgres, building support for Apache Iceberg tables (likely contributing to iceberg-rust), and wider object store coverage using Apache OpenDAL. Open-source under AGPL and contribution is encouraged through Slack community or giving a star on GitHub.