2026-09-17
每日一文 · 长文精读
Dropbox Evolves Riviera Content Processing Platform to Support AI Workloads
Dropbox 升级 Riviera 内容处理平台以支持 AI 工作负载
作者：Leela Kumili · InfoQ 原文
摘要：Dropbox将其Riviera平台从文件预览服务升级为通用内容处理平台，支持300多种文件格式和100多种转换能力，每秒执行数十万次转换。平台采用可组合的转换管道，分离编排与执行，并已用于AI处理。通过API暴露Markdown转换、音视频转录等功能，并与外部工作流编排系统集成。
Dropbox has evolved Riviera from a file preview service into a universal content processing platform supporting more than 300 file formats and more than 100 transformation capabilities.
Dropbox 已将 Riviera 从文件预览服务升级为通用内容处理平台，支持超过 300 种文件格式和超过 100 种转换能力。
The platform now performs hundreds of thousands of transformations per second and supports Search, Replay, Sign, and Dash, allowing AI workloads to reuse content processing infrastructure developed for earlier products.
该平台现在每秒执行数十万次转换，并支持 Search、Replay、Sign 和 Dash，使 AI 工作负载能够重用为早期产品开发的内容处理基础设施。
Riviera began as an internal service for generating previews across Dropbox's file formats.
Riviera 最初是作为一项内部服务，用于生成 Dropbox 各种文件格式的预览。
Instead of creating separate processing services for individual formats and outputs, engineers decomposed content processing into reusable transformations.
工程师没有为每种格式和输出创建独立的处理服务，而是将内容处理分解为可重用的转换。
A PowerPoint preview, for example, can be generated by converting the presentation to PDF and then converting PDF pages into images.
例如，PowerPoint 预览可以通过将演示文稿转换为 PDF，然后将 PDF 页面转换为图像来生成。
These transformations can then be composed into other processing pipelines.
这些转换随后可以组合到其他处理管道中。
The architecture separates orchestration from execution.
该架构将编排与执行分离。
A central component validates requests, composes transformation pipelines, dispatches work to backend workers, and manages caching.
一个中央组件验证请求、组合转换管道、将工作分派给后端工作进程并管理缓存。
Individual workers implement specific transformation capabilities, while a plugin model allows new capabilities to be added without changing the core orchestration layer.
各个工作进程实现特定的转换能力，而插件模型允许在不改变核心编排层的情况下添加新能力。
Riviera was already being used for AI processing before the latest expansion.
在最新扩展之前，Riviera 已用于 AI 处理。
Dropbox previously described chaining transformations to convert content into text and embeddings for AI-powered summaries and question answering.
Dropbox 之前描述了将转换链接起来，将内容转换为文本和嵌入，用于 AI 驱动的摘要和问答。
At that time, the platform supported roughly 300 file types and processed about 2.5 billion requests per day, representing nearly an exabyte of data.
当时，该平台支持大约 300 种文件类型，每天处理约 25 亿个请求，代表近艾字节的数据。
Cached intermediate results allowed multiple AI operations to reuse previously generated content.
缓存的中间结果允许多个 AI 操作重用先前生成的内容。
Dash introduced additional content preparation requirements because its search and AI capabilities process content from Dropbox and connected services.
Dash 引入了额外的内容准备需求，因为其搜索和 AI 能力处理来自 Dropbox 和连接服务的内容。
Riviera provides transformations for extracting and normalizing that content before indexing, allowing Dash to use the existing processing pipeline.
Riviera 提供了在索引之前提取和规范化该内容的转换，使 Dash 能够使用现有的处理管道。
Dropbox has also exposed Riviera capabilities through public APIs.
Dropbox 还通过公共 API 公开了 Riviera 的能力。
The developer documentation describes asynchronous endpoints for converting documents to Markdown, transcribing audio and video, and extracting structured metadata.
开发者文档描述了用于将文档转换为 Markdown、转录音频和视频以及提取结构化元数据的异步端点。
The Markdown API supports applications including indexing, rendering, and providing documents as input to large language models.
Markdown API 支持包括索引、渲染以及将文档作为大型语言模型输入在内的应用程序。
Applications submit jobs and poll using an asynchronous job identifier, with completion and failure states returned separately.
应用程序使用异步作业标识符提交作业并轮询，完成和失败状态分别返回。
The platform is also appearing in external developer tooling.
该平台也出现在外部开发者工具中。
A developer in the Dropbox Community published conductor-dropbox, an open source integration for Conductor OSS and Orkes that includes asynchronous Riviera Markdown extraction for document and RAG workflows.
Dropbox 社区的一位开发者发布了 conductor-dropbox，这是 Conductor OSS 和 Orkes 的一个开源集成，包括用于文档和 RAG 工作流的异步 Riviera Markdown 提取。
Riviera differs from general-purpose extraction frameworks such as Apache Tika, which provides APIs for detecting, parsing, and extracting text and metadata from a broad range of file formats.
Riviera 与 Apache Tika 等通用提取框架不同，后者提供用于检测、解析和从多种文件格式中提取文本和元数据的 API。
Riviera combines transformation plugins with pipeline composition, asynchronous execution, caching, and centralized orchestration.
Riviera 将转换插件与管道组合、异步执行、缓存和集中编排相结合。
Dropbox is now exposing selected Riviera capabilities through APIs while continuing to use the platform across document, multimedia, search, and AI workloads.
Dropbox 现在通过 API 公开选定的 Riviera 能力，同时继续在文档、多媒体、搜索和 AI 工作负载中使用该平台。
The external Conductor integration provides an example of Riviera's content extraction capabilities being incorporated into a broader workflow orchestration system.
外部的 Conductor 集成提供了一个示例，展示了 Riviera 的内容提取能力如何被纳入更广泛的工作流编排系统。
阅读理解
1. What was Riviera's original purpose?
2. How does Riviera handle generating a PowerPoint preview?
3. What additional benefit does Dash get from Riviera?