我们如何为 ClickHouse 构建一个新的强大的 JSON 数据类型

Update January 2025: Benchmarked ClickHouse's new JSON implementation against other data stores and provided results link.
Update March 2025: Demonstrated how to accelerate JSON queries for sub-100ms analytical performance regardless of data size.
JSON is the lingua franca for handling semi-structured and unstructured data.
ClickHouse recognized the importance of seamless JSON support but faced challenges.
ClickHouse is a fast analytical database with true column-oriented storage.
To enable high-performance for JSON data, implemented true column-oriented storage.
Challenges included handling different data types for the same JSON paths and avoiding an avalanche of column files.
Introduced the new JSON data type to address these challenges:
- Dynamically changing data with different data types.
- High-performance and dense true column-oriented storage.
- Scalability with limits on subcolumns.
- Tuning with hints for JSON parsing.
The Variant data type is a building block, allowing efficient storage of different data types.
The Dynamic type is an enhancement of the Variant type, with features like storing any data type and limiting types.
The new JSON type uses these building blocks and has optional parameters and hints.
It stores JSON objects with any structure and reads values using JSON paths as subcolumns.
Solves challenges like preventing an explosion of column files.
Supports reading leave values and subcolumns using special syntax.
Implemented a special compact format for discriminators serialization.
The new JSON type replaces the deprecated Object('json') data type and is experimental.
The JSON roadmap includes enhancements like using JSON key paths in primary keys.
The building blocks also pave the way for supporting other semi-structured types.
Users can contact ClickHouse Cloud support for private preview access to the new JSON data type.