跳转至

Furnace

Furnace is Foundry’s modern SQL warehousing engine, purpose-built for flexibility, performance, and extensibility. Furnace lets you run powerful, flexible SQL queries on your data, using the latest industry standards for compute and interoperability. With Furnace, you can run read/write SQL on top of Foundry compute, both natively inside Foundry and from your choice of third-party analytics tools. Furnace is engineered from first principles to support multiple compute engines behind a common syntax, providing future-proofing and resilience for your SQL workloads.

Furnace architecture

Furnace’s architecture decouples query parsing and planning from query execution. This separation is key to Furnace's flexibility and extensibility:

  • Query parsing & planning: All SQL queries are parsed and planned using a unified engine-agnostic layer. This layer leverages Apache Calcite to parse SQL into an abstract syntax tree (AST), validate it, and optimize the query, regardless of the eventual compute engine.
  • Compute engine abstraction: After planning, Furnace determines which compute engine should execute the query. The compute engine is responsible for running the query, reading data, and returning results.
  • Engine selection: By default, Furnace intelligently selects the best available engine for each query based on workload, compatibility, and performance. Engine selection is automatic and requires no configuration from users.
  • Unified user experience: Regardless of which compute engine is used or whether the query entry point is inside the Foundry platform or external, users interact with Furnace through the same APIs and interfaces (including Arrow Flight SQL), and benefit from consistent support for modern table formats like Iceberg.

This architecture diagram demonstrates the information flow for interactive Furnace queries:

Diagram showing the architecture of Furnace, including routing to different compute engines and routing of large and small results.

Supported compute engines

Furnace's compute engine abstraction allows Palantir to update and optimize supported engines over time.

Currently supported Furnace engines are:

Unified SQL experience

Furnace provides a unified SQL experience across all connection methods:

All connection methods benefit from Furnace's intelligent engine selection and performance optimizations, regardless of how you connect to Foundry.

中文翻译

Furnace

Furnace 是 Foundry 的现代 SQL 数据仓库引擎，专为灵活性、性能和可扩展性而设计。Furnace 让您能够利用最新的计算和互操作性行业标准，对数据运行强大且灵活的 SQL 查询。借助 Furnace，您可以在 Foundry 计算平台上运行读写 SQL 操作，既可以在 Foundry 内部原生执行，也可以通过您选择的第三方分析工具进行。Furnace 从底层原理出发进行设计，支持在统一语法背后运行多种计算引擎，为您的 SQL 工作负载提供面向未来的保障和弹性。

Furnace 架构

Furnace 的架构将查询解析与规划（query parsing and planning）从查询执行（query execution）中解耦。这种分离是 Furnace 灵活性和可扩展性的关键：

  • 查询解析与规划： 所有 SQL 查询均通过统一的引擎无关层进行解析和规划。该层利用 Apache Calcite 将 SQL 解析为抽象语法树（abstract syntax tree，AST），进行验证并优化查询，无论最终使用何种计算引擎。
  • 计算引擎抽象： 规划完成后，Furnace 会确定应由哪个计算引擎执行查询。计算引擎负责运行查询、读取数据并返回结果。
  • 引擎选择： 默认情况下，Furnace 会根据工作负载、兼容性和性能，智能地为每个查询选择最佳可用引擎。引擎选择是自动完成的，无需用户进行任何配置。
  • 统一用户体验： 无论使用哪种计算引擎，也无论查询入口是在 Foundry 平台内部还是外部，用户都通过相同的 API 和接口（包括 Arrow Flight SQL）与 Furnace 交互，并受益于对 Iceberg 等现代表格式的一致支持。

以下架构图展示了交互式 Furnace 查询的信息流：

显示 Furnace 架构的示意图，包括路由到不同计算引擎以及大结果和小结果的路由。

支持的计算引擎

Furnace 的计算引擎抽象允许 Palantir 随时间推移更新和优化支持的引擎。

目前支持的 Furnace 引擎包括：

统一 SQL 体验

Furnace 在所有连接方式上提供统一的 SQL 体验：

无论您如何连接到 Foundry，所有连接方式都能受益于 Furnace 的智能引擎选择和性能优化。