Aurora PostgreSQL queries Iceberg and Parquet data lakes directly via embedded DuckDB

AWS is removing a long-standing separation between transactional and analytical data. With this launch, one PostgreSQL query in Aurora can join live operational tables with Iceberg or Parquet data stored in S3. Previously, that required building pipelines to copy data between systems. The analytical engine is DuckDB, embedded directly in Aurora. Supported catalogs and storage include Glue Data Catalog, plain S3 and S3 Tables.

Swami Sivasubramanian, AWS VP of Agentic AI, framed it this way: "Operational databases and data lakes have always been separate worlds... That's changing today." He noted that because the feature builds on open-source DuckDB, future improvements to DuckDB will flow into Aurora.

The agentic angle is that agents increasingly need both current state (orders, tickets, sessions) and historical context (lake data) in one step. Removing ETL means fresher data and fewer moving parts for agent tools. Related launches the same day point the same way:

- Aurora serverless now scales in 16-ACU steps within a second, aimed at bursty agentic workloads.

- S3 Tables now support all Iceberg V3 data types.

Competitive context: Microsoft's Fabric pitch is similar, unifying operational and analytical data for Copilot. Snowflake and Databricks offer lakehouse federation from the analytics side. Embedding DuckDB, rather than building a proprietary engine, is a pragmatic open-source choice.

Caveats: performance on large lake scans from inside an OLTP database is unknown. Running heavy analytical queries on production Aurora clusters could compete with transactional traffic for resources. Cost and concurrency guidance will matter.