motherduck-load-data

Use CTAS or INSERT...SELECT statements with remote-read settings to load Parquet files from object storage into MotherDuck. This batch loading approach avoids inefficient row-by-row inserts and ensures safe data landing.

To ingest local files into MotherDuck, use native DuckDB client ingestion pathways. This method supports local files and local DuckDB databases, applying batch loading patterns rather than individual row inserts for optimal performance.

You can use the Postgres endpoint, but only as a thin-client remote-read route rather than a native DuckDB ingestion surface. This means it is not used for direct bulk loading operations into MotherDuck tables.

You need to use CREATE SECRET when loading data from object storage or HTTPS remote sources that require authentication. It provides the necessary access credentials for your CTAS or INSERT...SELECT workflows.

Row-by-row inserts cause significant performance bottlenecks during MotherDuck data ingestion. Preferring batch loading, Parquet-first movement, and CTAS patterns ensures efficient data transfer and maintains a production-friendly loading posture.

Validate data after loading into MotherDuck by running row counts and type checks on the raw landing tables. Once validated, promote the data into staging or modeling layers only after the load is confirmed correct.