dirigent-block-duckdb

The DuckDB engine of dirigent's sql family. Install it beside dirigent-block-sql and a sql connection whose url is duckdb:///warehouse.duckdb or duckdb:///:memory: works like any other: the same two blocks, the same fields, the same rules.

uv pip install dirigent-block-duckdb

It registers under the dirigent.sql.engines.v1 entry-point group, so the family finds it with no configuration, and a worker that never runs a duckdb:// url carries none of it -- the engine binary is larger than every other driver the family speaks to put together.

DuckDB has no async driver at all, so every call runs in a worker thread, and a deadline that passes interrupts the engine rather than leaving a statement running behind a step that has already failed.

Files, and what holds them in

A statement here reads and writes the parquet and csv files a run holds: read_parquet(:source) and COPY ... TO :target take a file the same way a WHERE clause takes a value, and a file:// storage URI in params arrives as the path duckdb opens.

The boundary is duckdb's own, not a check on the parameters. A session is opened, given the run's work directory and its local scratch space as its allowed_directories, and then closed around them: enable_external_access goes off, which is what makes those roots the only paths the engine will open, and lock_configuration goes on, which refuses the SET that would give any of it back. So a path written straight into the sql is refused the same way a parameter outside the run is, and a statement cannot LOAD or INSTALL a further extension to reach past the boundary. Spilled intermediates and duckdb's secret store are pointed inside the run's work directory for the same reason.

Where a parameter or a statement names an s3:// object, the session loads duckdb's httpfs extension and is given the endpoint, region, credential and addressing style of the connection the s3 scheme is configured from, and that scheme stays reachable beside the run's own directories. The extension is loaded, never installed, at run time; a bare install does it once:

python -c "import duckdb; duckdb.connect().execute('INSTALL httpfs')"

docs/sql.md is the family's home, and its DuckDB section is the worked example.