Quack w/ Hannes Mühleisen (DuckDB)

TL;DR

• 1

  Quack gives DuckDB a native client-server protocol: Hannes framed it as DuckDB finally being able to talk to another DuckDB instance, whether on the same machine, across the network, or even from the browser via HTTP.

• 2

  The benchmark claim is aggressive and specific: In Hannes' test on two EC2 machines in the same availability zone, Quack transferred 16 million rows in about 4 seconds, versus 22 seconds for Arrow Flight and about 3 minutes for Postgres.

• 3

  Quack also surprised on transactional inserts: DuckDB was expected to shine at bulk transfer, but Hannes said Quack even outperformed Postgres on a small-transaction benchmark with eight parallel threads.

• 4

  The protocol is intentionally boring in the right places: Quack runs over TCP plus HTTP, not a custom transport, so it works with browsers, proxies, firewalls, and standard ops tooling like Nginx.

• 5

  Authentication and authorization are pluggable by design: Quack ships with basic token auth, but teams can override auth and query authorization with extensions, including community-built OAuth support and custom SQL functions.

• 6

  The most interesting near-term use case may be DuckLake catalogs: Hannes showed DuckDB serving as a remote catalog server for DuckLake through Quack, which matters because DuckLake's inlining path for small inserts was already hitting limits with Postgres.