Support read_parquet and similar DuckDB file I/O functions directly in Postgres views

[visardida]

Enable the ability to create persistent views or tables in PostgreSQL that reference external files via DuckDB’s read_parquet, read_csv, or other file readers — without requiring ingestion of the data into Postgres.

Currently, pg_duckdb allows embedding DuckDB as an engine inside PostgreSQL, but to query external files you must use transient DuckDB runtime calls, not schema-persistent objects.

Supporting file-backed views would allow PostgreSQL users to:
• Instantly query large Parquet datasets without ETL or import.
• Persist views that reference DuckDB’s file readers (read_parquet, read_csv, read_json, etc.).
• Integrate seamlessly with BI tools and ORMs that expect standard Postgres tables or views.
• Use pg_duckdb as a lightweight analytical layer for external data lakes — a true “zero-ETL” analytics pattern.

Example:

CREATE VIEW my_view AS
SELECT *
FROM read_parquet(
  '/path/to/data/**/*.parquet',
  union_by_name=1
);

SELECT * FROM my_view LIMIT 10;

Expected Behavior
• PostgreSQL can store the view definition normally (pg_views, pg_catalog).
• On query execution, pg_duckdb should delegate the underlying execution plan to DuckDB.
• The DuckDB engine should lazily read the files at runtime, respecting the view definition.
• The user can refresh or drop the view like any other Postgres object.
• This feature should work in combination with DuckDB’s file globbing (**/*.parquet) and options like union_by_name.

Benefits
• Zero-copy, zero-ETL analytics on file-based datasets.
• Perfect for hybrid data lake / Postgres integration scenarios.
• Extremely fast development iteration (no loading, immediate schema exposure).
• Brings pg_duckdb to parity with the most useful part of MyDuckServer’s feature set.

[YuweiXiao]

Hi, we have a PoC feature called duckdb external table with the following grammar:

CREATE TABLE external_parquet () USING duckdb WITH (
    duckdb_external_location = '../../data/iris.parquet',
    duckdb_external_reader = 'read_parquet'
);

select * from external_parquet;

The table columns are inferred automatically from the data files. Will post a link here once the PR is drafted.

[visardida]

Fantastic, very much looking forward to this. Thank you!

[YuweiXiao]

Hi! Here's the PR link. Feedback is welcome!