Skip to main content

A Flight SQL server for DuckDB: JDBC, ODBC and ADBC

DuckDB does not ship a server: it is a library, and its bindings open a database inside your process. Quack on Demand is an Arrow Flight SQL server for DuckDB and DuckLake, which means every tool that carries a Flight SQL driver, JDBC, ODBC or ADBC, connects to shared DuckDB data the way it connects to any warehouse.

The problem​

BI tools, SQL editors and services on other machines need a network endpoint and a standard driver. Serving a DuckDB file over HTTP or a filesystem share gives you neither: each client still runs its own engine, sees a stale or locked file, and carries no identity. You want one endpoint, one identity model, and drivers your tools already have.

How Quack on Demand does it​

One command boots the manager, the Flight SQL edge on :31338, DuckDB's native Quack listener on :9494 and the admin UI, then prints ready-to-paste connection strings for every driver family:

uvx qod@latest serve ./sales.duckdb
#   JDBC : jdbc:arrow-flight-sql://localhost:31338/?tenant=<tenant>&pool=<pool>&user=<user>&useEncryption=true&disableCertificateVerification=true
#   ADBC : uri=grpc+tls://localhost:31338  (adbc_driver_flightsql)
#   ODBC : Driver={Arrow Flight SQL ODBC Driver};Host=localhost;Port=31338;...

The JDBC driver is the Apache Arrow Flight SQL JDBC driver from Maven Central, which DBeaver, Tableau and Spark load directly. ADBC uses the Flight SQL ADBC driver (adbc_driver_flightsql), which Power BI also ships in-box. ODBC uses any third-party Flight SQL ODBC driver. Behind the edge, statements are classified read or write and routed to the least-loaded DuckDB node in the pool, with transactions pinned to one node. TLS is on by default. Quack on Demand runs as a single Docker container on one node, or on Kubernetes for multi-host fleets.

Go deeper​