All Posts

Data & Analytics

How DuckDB 2.0's Quack Server Mode Works and What It Changes

How DuckDB 2.0's Quack Server Mode Works and What It Changes

TechShorts Newsroom

Editorial Team

DuckDB started as a database that runs inside another program, with no server to install or manage. Version 2.0.0, released on 21 October 2026, adds an optional one. The announcement calls 2.0 "above all a feature release" and designates it a long-term support (LTS) version, supported until at least October 2027.

The headline change, and the focus of AlternativeTo's coverage, is Quack, a client-server protocol that lets one DuckDB process share its databases with other processes over the network.

The limit Quack addresses

An embedded database runs in the same process as the application using it. That keeps deployment simple but limits who can write. The concurrency documentation describes two modes: one process opens a database file with read and write access, or several processes open it read-only. Before Quack, writing to DuckDB's native file format from multiple processes was not supported.

The limit bites when a scheduled job and a notebook need the same file, because while one of them has the file open for writing, the other cannot open it. The project's own recommended workaround was DuckLake, its lake format, with PostgreSQL acting as a shared catalog. The other route was a hosted service built on DuckDB.

How Quack works

Quack runs over HTTP or HTTPS. One DuckDB process starts a server and exposes its databases; other DuckDB instances attach the remote endpoint much as they would a local file. A new CONNECT statement then routes the session's queries to the remote database and returns the results, according to the 2.0 CONNECT documentation.

The DuckDB team published a technical preview of Quack on 12 May 2026, during the 1.5 release series. The 2.0 concurrency page describes the shipped version as "polished", with performance and stability improvements, and says writing to the native format from multiple processes is now supported through Quack.

What "multi-process writes" means here is narrower than it sounds. Clients do not write to the file themselves; they send queries to the serving process, which owns it. Inside that process, DuckDB still uses multi-version concurrency control with optimistic conflict detection. Concurrent writes succeed when they do not conflict. When two writers change the same row at the same time, the second fails with a conflict error, per the concurrency documentation.

CONNECT also reaches beyond DuckDB. The announcement says the statement can target PostgreSQL and MySQL, and that a remote pushdown optimizer ships queries to the remote server instead of pulling tables over the wire. The CONNECT documentation notes that in-process databases, including the default in-memory database and local DuckDB files, do not support it.

Three ways to share a DuckDB database

Quack is now one of three main routes, and each puts the operational work in a different place.

A Quack server keeps the existing DuckDB file and puts one process in charge of it. It suits an internal application or a small team that wants several clients on one database without adopting new storage. Every query against the shared database passes through the serving process, so the machine running it sets the limit on capacity.

DuckLake with a PostgreSQL catalog lets many DuckDB instances read and write concurrently by coordinating through a central catalog database. The 2.0 concurrency page still lists it as the alternative to Quack. It avoids a single serving process, at the cost of a second system to run and secure.

A managed service built on DuckDB, such as MotherDuck, runs the infrastructure for the customer. Quack gives open-source users their own route to shared remote access, but the team running the server takes on its security and upgrades.

What to check before running a Quack server

According to the Quack documentation, the server binds only to localhost by default and generates a random authentication token at startup unless one is supplied. Clients present the token through a secret scoped to the server's URI or an explicit TOKEN option.

Authentication is not authorization. The Quack security documentation says a server exposes the full SQL surface of the DuckDB instance behind it, including read and write access to every table its session can see. The default authorization callback is permissive and passes every query through unchanged, so any client with a valid token gets the same access as any other. Both callbacks can be replaced, for example with SQL macros that check a table of per-user tokens or admit only read queries. The documentation advises against exposing Quack directly to the internet and recommends a reverse proxy such as nginx to terminate TLS.

Governance is the other check. DuckDB 2.0 is the first major release since Amazon Web Services agreed to acquire DuckLabs, the Amsterdam-based company behind DuckDB, a deal announced on 26 August 2026 and reported by SiliconANGLE the same day. Amazon said it was not acquiring the open-source project, which stays under the independent DuckDB Foundation and the MIT license. It also said the DuckLabs team would join AWS when the deal closes, with DuckDB's creators continuing to lead the project's technical direction.

What commonly goes wrong in the upgrade

Storage is the main trap. The 2.0 migration guide says new database files default to storage version 69, which DuckDB versions before 2.0.0 cannot read. Existing files keep their storage version when 2.0 writes to them, so they stay readable by older releases. The guide warns specifically about deployments where processes run different versions: a job upgraded to 2.0 can create a new file that an older client library elsewhere, in a notebook or BI connector, cannot open.

Teams that need older readers to keep working can write the previous format with the storage_compatibility_version setting or the STORAGE_VERSION option on ATTACH. To convert existing files, the guide describes exporting with the old binary and importing with 2.0, or using COPY FROM DATABASE.

The same guide lists smaller breaking changes. The old single-arrow lambda syntax is now disabled by default; a session setting re-enables it, but DuckDB 2.1 will remove both. Integer division and modulo by zero now raise an error instead of returning NULL, and the ICU library is no longer bundled, which matters only for custom builds.

The engine changes also touch everyday queries. The announcement lists asynchronous I/O for reading Parquet, CSV and native database files, full integration of the VARIANT type introduced in 1.5, and a PEG-based parser as the new default. The new parser is designed to accept what the old one did, but the migration guide lists a few expressions that deliberately behave differently and can silently change results. The ~ regular expression operator, for example, now performs partial matching, following PostgreSQL. That is a reason to run existing SQL through regression tests before moving production jobs.

The deadline to watch is on the previous LTS line. Both the announcement and the migration guide give DuckDB 1.4 LTS an end-of-life date of 17 November 2026. Support for 1.5 has been extended to March 2027. For mission-critical workloads, the announcement suggests waiting for the first 2.0 patch release, scheduled for November.

Sources

  • DuckDB. "Announcing DuckDB 2.0.0 LTS 'Cinnamon Teal'." 21 October 2026. https://duckdb.org/2026/10/21/announcing-duckdb-200
  • AlternativeTo. "DuckDB's major 2.0 release turns the database management system into a full-fledged server." October 2026. https://alternativeto.net/news/2026/10/duckdb-s-major-2-0-release-turns-the-database-management-system-into-a-full-fledged-server/
  • DuckDB. "Quack: The DuckDB Client-Server Protocol." 12 May 2026. https://duckdb.org/2026/05/12/quack-remote-protocol.html
  • DuckDB documentation. Concurrency page, current stable version. 2026. https://duckdb.org/docs/current/connect/concurrency
  • DuckDB documentation. Concurrency page, version 2.0. 2026. https://duckdb.org/docs/2.0/connect/concurrency
  • DuckDB documentation. CONNECT statement, version 2.0. 2026. https://duckdb.org/docs/2.0/sql/statements/connect
  • DuckDB documentation. Quack overview. 2026. https://duckdb.org/docs/current/quack/overview.html
  • DuckDB documentation. Quack security, version 2.0. 2026. https://duckdb.org/docs/2.0/quack/security
  • DuckDB documentation. Migration guide, version 2.0. 2026. https://duckdb.org/docs/2.0/operations_manual/migration
  • Amazon. Announcement of the agreement to acquire DuckLabs. 26 August 2026. https://aboutamazon.com/news/company-news/aws-ducklabs
  • SiliconANGLE. Report on AWS buying DuckLabs. 26 August 2026. https://siliconangle.com/2026/08/26/aws-buys-ducklabs-to-bring-duckdbs-embeddable-analytics-to-more-enterprises/