DuckDB

DuckDB was developed by the CWI database research center in the Netherlands and is positioned as the SQLite of the OLAP domain, focusing on columnar storage and a vectorized execution engine

DuckDB是近年来数据工程领域增长最快的开源项目之一，是一个嵌入式的列式分析数据库

DuckDB 2.0 includes the built-in Quark protocol and a new Connect statement, which can connect to remote DuckDB as well as directly to Postgres SQL and MySQL and push queries down to them for execution.

DuckDB 2.0 brings asynchronous reading of Pocket and CSV, a genuinely usable Variant semi-structured type, and triggers.

The DuckDB 2.0 command line can recognize that it is being called by an Agent and automatically switch to concise output, which in tests reduced the content the model had to read by nearly 60%.

The emergence of 2.0.0 LTS shifts the competitive dimension from "who is faster" to "who is more trustworthy," and in production environments the predictability of version upgrades will outweigh millisecond differences in benchmarks.

DuckDB is positioned as in-process OLAP, requiring no standalone service or network connection, and is embedded directly in applications as a library.

DuckDB and cluster-deployment solutions like ClickHouse and Doris are complementary rather than substitutes.

Teams already using DuckDB in production should add 2.0.0 LTS to their upgrade evaluation plans and first run regression tests with existing workloads, focusing on the compatibility of extensions (httpfs, parquet, iceberg) and custom UDFs.

DuckDB has been known over the past few years for its extremely fast iteration pace, with new features emerging almost every month.

For individual users who treat DuckDB as a temporary analysis scripting tool, the urgency of upgrading to 2.0.0 LTS is not high.

DuckDB jumped directly from 1.x to version number 2.0.0, which under semantic versioning conventions means public interfaces may undergo breaking changes.

DuckDB long used 0.x version numbers before 1.0, and its version semantics differ from strict SemVer; the actual scope of changes should still be based on the official release notes, and the impact cannot be inferred from version numbers alone.

DuckDB is an embedded analytical (OLAP) database written in C++, known for being single-file and zero-configuration.

DuckDB is designed mainly around a single writer, and multiple processes writing to the same file simultaneously is restricted, making it unsuitable for high-concurrency online services.

DuckDB natively supports directly querying formats such as Parquet, CSV and JSON.

croft is an open-source tool that lets users write data pipelines in TypeScript and store all data in a single DuckDB file.

croft's core design is to pack all assets of a data pipeline (raw data, transformation logic, scheduling state) into a single DuckDB file.

DuckDB uses columnar storage and a vectorized execution engine, and its performance on analytical queries involving aggregation, joins and scans of large tables is significantly better than that of traditional row-store transactional databases.

In DuckDB, table functions such as read_csv, read_parquet and glob look like SELECT but can actually read local or remote files (such as /etc/passwd), forming a filesystem side channel; they must be blocked by function name, and literal table references such as FROM 'data.csv' must also be intercepted.