This page tracks the documentation for the features introduced in DuckDB 2.0, organized around the themes of the DuckDB 2.0.0 release announcement. Each section links to the relevant documentation.

DuckDB as a Server: Quack and CONNECT

Since v1.5.3, DuckDB can run as a server via the Quack protocol, letting multiple clients connect to a shared instance over the network. DuckDB 2.0 ships a polished version of the quack extension, which provides the remote protocol, and adds the new CONNECT and DISCONNECT statements to open and close connections from SQL. The command line client can start a server and attach to one directly.

• Quack extension
• CONNECT and DISCONNECT statements
• CLI -serve and -connect arguments
• Securing Quack with a reverse proxy
• Remote pushdown to PostgreSQL and MySQL

VARIANT Becomes a First-Class Citizen

The VARIANT type stores semi-structured, schema-less data such as JSON while keeping the performance of columnar execution. A family of variant_* functions constructs and inspects variant values.

• Variant type
• Variant functions

Triggers

DuckDB 2.0 added trigger support, so you can run actions automatically in response to table changes. This is useful for building audit tables and enforcing invariants.

• CREATE TRIGGER statement

SQL Dialect Additions

DuckDB 2.0 extended the SQL dialect in several directions, from new join types to data-modifying CTEs and a friendlier variable syntax.

• Explicit join type syntax, including nearest (NEAREST) joins
• Data-modifying CTEs, including COPY ... blocks in CTEs
• Nested schemas
• The $variable variable shorthand, including session variables as default parameter values
• JSON mutation functions
• Grouping by UNNEST
• The SQL-standard FETCH FIRST syntax
• Well-defined semantics for multiple matching source rows in MERGE INTO and UPDATE ... FROM
• SECURE VIEW
• SHOW statement
• Setting a sequence value
• regexp_extract additions

Asynchronous I/O

DuckDB 2.0 issues reads in the background and keeps multiple requests in flight, which speeds up queries that read from object storage such as S3. The behavior is controlled by the async_threads and read-ahead settings.

• Asynchronous I/O overview
• Thread pools and read-ahead
• Asynchronous I/O configuration settings

Faster Queries Across the Board

DuckDB 2.0 included optimizer and engine improvements, including faster recursive CTEs and better row-group pruning.

• Recursive CTEs and the blog post on how DuckDB runs them faster
• Min-max indexes (zonemaps) used for row-group pruning

Storage Format v2.0

New database files use the v2.0.0 storage format by default. The format adds lazy metadata loading, better compression and corruption validation. DuckDB versions before 2.0 cannot read files in the new format, so a migration guide covers upgrading existing deployments.

• Default storage version
• Storage version table
• Migrating existing deployments: see the DuckDB 2.0 migration guide

A Brand New SQL Parser

DuckDB 2.0 introduced a PEG-based SQL parser that replaces the PostgreSQL-derived one and allows extensions to contribute grammar.

• PEG parser
• Legacy parser extension for switching back to the old parser

Timezones, Calendars and Collations Without ICU

DuckDB 2.0 replaces the ICU library with a smaller, faster native implementation of timezone, calendar and collation support in the icu extension.

• ICU extension

Write Extensions Once, Host Them Yourself

DuckDB 2.0 shipped a revamped C API (C API v2), which is not yet stable. You can also register and host your own extension repositories or statically link extensions into custom builds.

• C API v2
• Extension repositories
• Statically linking extensions: see Statically Linking Extensions

Core Extension Updates

Several core extensions shipped new features alongside DuckDB 2.0.

• DuckLake extension, the reference implementation of DuckLake 1.1
• Iceberg extension ATTACH options for REST catalogs
• Creating Delta tables, time travel and VARIANT columns in the Delta extension

Deprecations

DuckDB 2.0 disables or phases out a few features, which will be removed in future releases.

• The single-arrow lambda syntax, now disabled by default
• macOS x86_64 (osx_amd64), for which community extensions are no longer built

Other Documentation Updates

Additional 2.0 features documented across the site, grouped by area.

SQL

• TIMESTAMPTZ_NS, time zone aware timestamps with nanosecond precision
• The lttb aggregate for downsampling time series data
• The new extension_name column in duckdb_functions
• The duckdb_secret_type_parameters() function listing the parameters each secret type accepts
• The is_generated and generation_expression columns in duckdb_columns() and information_schema.columns
• The decimal_division function for exact decimal division
• The decimal_average aggregate for exact decimal averages
• The variant_group_object aggregate for building VARIANT objects
• The variant_group_array function for collecting values into a VARIANT array
• Renaming a view with ALTER TABLE ... RENAME TO for PostgreSQL compatibility
• EXCEPT as a synonym for EXCLUDE in star expressions
• No-argument GROUPING_ID() and GROUPING() defaulting to all GROUP BY columns
• Backtick-quoted identifiers as an alternative to double quotes
• EXPLAIN (SQL) rendering the optimized plan as executable SQL
• Empty STRUCT types

Data Import and Export

• Writing GeoJSON and loading GeoJSON
• The Parquet schema parameter and new Parquet writing options
• The PARTITION_PATH copy option for computing custom partition subdirectories
• The ORDER_BY copy option for sorting rows before writing
• The TIMESTAMP_IS_ADJUSTED_TO_UTC and WRITE_TIMESTAMP_AS_INT96 Parquet writing options for controlling how timestamps are written
• The file_row_group_number virtual column for Parquet row group pruning
• The rejects_line_size_limit CSV parameter for bounding stored rejected lines

Clients

• CLI agent mode, which produces compact, machine-readable output when the CLI is driven by an AI coding agent
• CLI command history: the .history command and the shell_history() table function
• Pretty-printed, highlighted .schema output with --no-indent to restore the single-line form
• Migrating Python code to DuckDB 2.0: the changes that code written for DuckDB 1.x needs
• Python relations that run their query on every read, and parallel Python queries with one cursor per thread
• Open results of the Python DB API with complete() and abort(), and Python transactions that a failed execute() aborts
• The Python appender for inserting rows that Python code produces
• Python expression helpers such as col(), lit() and fn(), aggregates and window functions on expressions and new relation methods such as with_columns(), group_by() and pivot() in the Relational API

Configuration and Profiling

• Setting max_memory as a percentage of total system memory
• The max_execution_time setting for bounding query wall-clock time
• The checkpoint_on_detach setting controlling checkpoints when detaching a database
• Deprecation of implicit identifier-to-string conversion in table function arguments, with the table_function_identifier_conversion setting
• Removal of the enable_http_logging and http_logging_output settings in favor of the unified logging framework
• The http_client_pool_capacity setting for sizing the asynchronous HTTP client pool
• The metrics layer and the tracked_metrics setting
• Integer division by zero now raises an error
• The web profiling format and the CLI .web command for opening query graphs in a browser

Building

• A C++17 compiler is now required to build DuckDB