AWS says Amazon Aurora PostgreSQL can query operational records alongside Apache Iceberg tables and Apache Parquet data in Amazon S3 in a single query. DuckDB is embedded in Aurora to process analytical scans, so applications can combine live records—including uncommitted writes—with lake data without first copying it into Aurora through an ETL pipeline.

Aurora PostgreSQL queries lake data with operational records

The feature brings data from Aurora’s operational database together with Iceberg and Parquet data stored in S3, including S3 Tables. That lets an application query current records and data in the lake through PostgreSQL, rather than first moving the lake data into Aurora.

AWS describes DuckDB as the analytical engine embedded in Aurora for those scans. This is Aurora’s managed feature; pg_duckdb is a separate PostgreSQL extension with its own deployment model.

Supported versions and catalog setup

AWS lists support beginning with Aurora PostgreSQL 17.11 and 18.6. Enabling the feature requires the aurora_analytics extension and an IAM role configured with the AuroraAnalytics feature. The role gives Aurora access to S3 and AWS Glue.

For compatible external Iceberg REST Catalogs, AWS’s setup uses AWS Glue Data Catalog to register the catalog, then foreign tables that reference its data. This also supports Iceberg tables managed through Glue.

Query behavior and materialization

AWS describes predicate pushdown, column pruning and caching as query optimizations. The aurora_analytics_stat_statements() function reports rows scanned, S3 bytes read and cache hits, giving teams metrics to inspect for individual queries.

Read queries may run on the Aurora writer or a read replica. Commands that materialize selected lake data into native Aurora tables run on the writer. For workloads that need single-digit-millisecond latency, AWS points to materialization as an option.

In one AWS example, Aurora combines seven days of recent transactions with five years of historical transactions in an S3 Parquet file. Aurora infers the foreign-table schema from the file metadata in that example.

Availability and charges

AWS says the feature is available in all commercial AWS Regions and AWS GovCloud (US) Regions. There is no additional feature charge; incremental Aurora compute and S3 request charges apply.