Amazon Aurora's analytics is DuckDB: a reproducible side-by-side with pg_duckdb

Microsoft unveiled the pg_duckdb extension for Azure Database for PostgreSQL Flexible Server at the Ignite conference on November 18, 2025. Amazon announced their intention to acquire DuckLabs, the creators of DuckDB, in August 2026; however, the open-source DuckDB project remains independent. On September 30, 2025, Amazon announced that Amazon Aurora PostgreSQL can utilize embedded DuckDB to query data in Apache Iceberg and Parquet formats stored in data lakes alongside current data.

Amazon Aurora PostgreSQL can now query Parquet and Iceberg files in S3 through the aurora_analytics extension, and AWS claims that DuckDB is embedded within it. To demonstrate this, I compared Azure and Amazon's execution plans for the same query over the same file, using the open-source pg_duckdb engine.

To set up the comparison, I provisioned an Aurora PostgreSQL cluster with aurora_analytics support, requiring a supported PostgreSQL version (17.11+ or 18.6+) and an IAM role with the AuroraAnalytics feature to read from the S3 bucket. I utilized a CloudFormation stack for this setup, including an Aurora PostgreSQL Serverless v2 cluster, an engine version parameter, an aurora_analytics-enabled cluster parameter group, an IAM role granting S3 read access to the data bucket (and Glue read for Iceberg), and an S3 bucket for the Parquet data.

After uploading a Parquet file to the S3 bucket and connecting to Aurora, I enabled the aurora_analytics extension and created a foreign table over the file. The query executed was a SELECT statement to count records grouped by production_year from the parquet file.

The query and plan for Amazon Aurora PostgreSQL were captured with the EXPLAIN (ANALYZE, VERBOSE) command. The plan showed a Custom Scan, Output projection, Pushdown SQL, and various PROJECTION steps, including decompression and compression of integral data types, perfect hash group-by operations, and final ordering by the count in descending order.

By comparing the execution plans from Aurora and pg_duckdb, it is evident that both systems process the query in a similar manner, confirming Amazon's claim that DuckDB is embedded within Amazon Aurora PostgreSQL.