MD_RUN parameter

For certain DuckDB Table Functions, MotherDuck now provides an additional parameter, MD_RUN that gives explicit control over where the query is executed.

This parameter is available to the following functions:

• read_csv()
• read_csv_auto()
• read_json()
• read_json_auto()
• read_parquet() and its alias parquet_scan()

To leverage the MD_RUN parameter, you can choose:

• MD_RUN=LOCAL executes the function in your local DuckDB environment
• MD_RUN=REMOTE executes the function in MotherDuck-hosted DuckDB runtimes in the cloud
• MD_RUN=AUTO executes remotely all s3://, http://, and https:// requests, except those to localhost/127.0.01. This is the default option.

The following is an example of evoking this parameter to execute the function remotely:

SELECT *
FROM read_csv_auto(
    'https://github.com/duckdb/duckdb/raw/main/data/csv/ips.csv.gz',
    MD_RUN=REMOTE)
LIMIT 100

In this example MD_RUN=REMOTE is redundant, because omitting it implies MD_RUN=AUTO and given that this is a non-local https:// resource, MotherDuck will automatically chose remote execution already.

One can force local execution with MD_RUN=LOCAL. Be aware that DuckDB-WASM does not support reading compressed files yet, so inside the Web Browser one would get an error for this particular file as it is ips.csv**.gz** (it does work locally from the CLI or e.g. a python notebook).