RAG核心特性详解：ETL流水线

ETL流水线

前言

在 RAG（检索增强生成）应用中，有一个绕不开的核心环节——如何把海量的文档"喂"给 AI？这些文档可能散落在各处：有的是 PDF 报告，有的是 Markdown 笔记，有的是飞书文档，甚至还有 B 站视频字幕。直接把它们扔给大模型？不现实，大模型的上下文窗口有限，而且一股脑塞进去效果也很差。

这就引出了今天的主角：ETL Pipeline。

ETL 是三个单词的缩写：Extract（抽取）、Transform（转换）、Load（加载），它是 Spring AI 中处理文档流水线的核心框架。简单来说，ETL 做了一件事：把原始文档变成向量数据库中可检索的知识碎片。

读完这篇博客，你将彻底搞懂：

• ETL 到底是什么，每个阶段在干什么
• Spring AI 提供了哪些开箱即用的"武器"
• Spring AI Alibaba 社区又补充了哪些"神器"
• 如何用代码实现一个完整的 ETL 流水线

一、大白话理解 ETL

先抛开代码，用生活场景来理解 ETL。

想象你要开一家"AI 智能客服"，知识来源是一堆乱七八糟的资料：PDF 产品手册、Word 操作指南、公司内部网页、Excel 价目表……

直接把这些资料甩给 AI 说"你学一下"，显然不行。你得先做三件事：

1. Extract（抽取）——“把书从书架拿下来”

首先你得能"读取"各种格式的文档。PDF 需要 PDF 阅读器，Word 需要 Word 解析器，网页需要 HTML 爬虫——这就是 DocumentReader 干的活。它负责从各种数据源把文档读进内存。

2. Transform（转换）——“把书拆成卡片”

一本 500 页的手册不可能一口气喂给 AI（也喂不下）。你得把它拆成一小段一小段的"知识卡片"，每张卡片包含一个相对完整的信息点。可能还要给卡片打标签、写摘要——这就是 DocumentTransformer 干的活。

3. Load（加载）——“把卡片放进柜子”

拆分好的知识卡片得存到某个地方，方便以后快速查找。RAG 场景下通常是存到向量数据库（Vector Store），把每张卡片转成向量——这就是 DocumentWriter 干的活。

用一张流程图来概括：

#mermaid-svg-st9CDcotGL4IOVYW{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-st9CDcotGL4IOVYW .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-st9CDcotGL4IOVYW .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-st9CDcotGL4IOVYW .error-icon{fill:#552222;}#mermaid-svg-st9CDcotGL4IOVYW .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-st9CDcotGL4IOVYW .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-st9CDcotGL4IOVYW .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-st9CDcotGL4IOVYW .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-st9CDcotGL4IOVYW .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-st9CDcotGL4IOVYW .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-st9CDcotGL4IOVYW .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-st9CDcotGL4IOVYW .marker{fill:#333333;stroke:#333333;}#mermaid-svg-st9CDcotGL4IOVYW .marker.cross{stroke:#333333;}#mermaid-svg-st9CDcotGL4IOVYW svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-st9CDcotGL4IOVYW p{margin:0;}#mermaid-svg-st9CDcotGL4IOVYW .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-st9CDcotGL4IOVYW .cluster-label text{fill:#333;}#mermaid-svg-st9CDcotGL4IOVYW .cluster-label span{color:#333;}#mermaid-svg-st9CDcotGL4IOVYW .cluster-label span p{background-color:transparent;}#mermaid-svg-st9CDcotGL4IOVYW .label text,#mermaid-svg-st9CDcotGL4IOVYW span{fill:#333;color:#333;}#mermaid-svg-st9CDcotGL4IOVYW .node rect,#mermaid-svg-st9CDcotGL4IOVYW .node circle,#mermaid-svg-st9CDcotGL4IOVYW .node ellipse,#mermaid-svg-st9CDcotGL4IOVYW .node polygon,#mermaid-svg-st9CDcotGL4IOVYW .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-st9CDcotGL4IOVYW .rough-node .label text,#mermaid-svg-st9CDcotGL4IOVYW .node .label text,#mermaid-svg-st9CDcotGL4IOVYW .image-shape .label,#mermaid-svg-st9CDcotGL4IOVYW .icon-shape .label{text-anchor:middle;}#mermaid-svg-st9CDcotGL4IOVYW .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-st9CDcotGL4IOVYW .rough-node .label,#mermaid-svg-st9CDcotGL4IOVYW .node .label,#mermaid-svg-st9CDcotGL4IOVYW .image-shape .label,#mermaid-svg-st9CDcotGL4IOVYW .icon-shape .label{text-align:center;}#mermaid-svg-st9CDcotGL4IOVYW .node.clickable{cursor:pointer;}#mermaid-svg-st9CDcotGL4IOVYW .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-st9CDcotGL4IOVYW .arrowheadPath{fill:#333333;}#mermaid-svg-st9CDcotGL4IOVYW .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-st9CDcotGL4IOVYW .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-st9CDcotGL4IOVYW .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-st9CDcotGL4IOVYW .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-st9CDcotGL4IOVYW .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-st9CDcotGL4IOVYW .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-st9CDcotGL4IOVYW .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-st9CDcotGL4IOVYW .cluster text{fill:#333;}#mermaid-svg-st9CDcotGL4IOVYW .cluster span{color:#333;}#mermaid-svg-st9CDcotGL4IOVYW div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-st9CDcotGL4IOVYW .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-st9CDcotGL4IOVYW rect.text{fill:none;stroke-width:0;}#mermaid-svg-st9CDcotGL4IOVYW .icon-shape,#mermaid-svg-st9CDcotGL4IOVYW .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-st9CDcotGL4IOVYW .icon-shape p,#mermaid-svg-st9CDcotGL4IOVYW .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-st9CDcotGL4IOVYW .icon-shape .label rect,#mermaid-svg-st9CDcotGL4IOVYW .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-st9CDcotGL4IOVYW .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-st9CDcotGL4IOVYW .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-st9CDcotGL4IOVYW :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

原始文档PDF/Word/MD/HTML…

ExtractDocumentReader读取文档

TransformDocumentTransformer分割、增强、格式化

LoadDocumentWriter写入向量数据库

向量数据库Vector Store

AI 检索增强生成

用一句大白话总结：ETL 就是"读取 → 拆分加工 → 存入向量库"的流水线，是 RAG 系统的数据入口。

二、Spring AI ETL 三大核心接口

Spring AI 用三个极简的 Java 接口定义了 ETL 的三个阶段，每个接口都用到了 Java 函数式编程的思想，非常优雅。

#mermaid-svg-jV25FQZBShFgosZy{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-jV25FQZBShFgosZy .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-jV25FQZBShFgosZy .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-jV25FQZBShFgosZy .error-icon{fill:#552222;}#mermaid-svg-jV25FQZBShFgosZy .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-jV25FQZBShFgosZy .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-jV25FQZBShFgosZy .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-jV25FQZBShFgosZy .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-jV25FQZBShFgosZy .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-jV25FQZBShFgosZy .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-jV25FQZBShFgosZy .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-jV25FQZBShFgosZy .marker{fill:#333333;stroke:#333333;}#mermaid-svg-jV25FQZBShFgosZy .marker.cross{stroke:#333333;}#mermaid-svg-jV25FQZBShFgosZy svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-jV25FQZBShFgosZy p{margin:0;}#mermaid-svg-jV25FQZBShFgosZy .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-jV25FQZBShFgosZy .cluster-label text{fill:#333;}#mermaid-svg-jV25FQZBShFgosZy .cluster-label span{color:#333;}#mermaid-svg-jV25FQZBShFgosZy .cluster-label span p{background-color:transparent;}#mermaid-svg-jV25FQZBShFgosZy .label text,#mermaid-svg-jV25FQZBShFgosZy span{fill:#333;color:#333;}#mermaid-svg-jV25FQZBShFgosZy .node rect,#mermaid-svg-jV25FQZBShFgosZy .node circle,#mermaid-svg-jV25FQZBShFgosZy .node ellipse,#mermaid-svg-jV25FQZBShFgosZy .node polygon,#mermaid-svg-jV25FQZBShFgosZy .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-jV25FQZBShFgosZy .rough-node .label text,#mermaid-svg-jV25FQZBShFgosZy .node .label text,#mermaid-svg-jV25FQZBShFgosZy .image-shape .label,#mermaid-svg-jV25FQZBShFgosZy .icon-shape .label{text-anchor:middle;}#mermaid-svg-jV25FQZBShFgosZy .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-jV25FQZBShFgosZy .rough-node .label,#mermaid-svg-jV25FQZBShFgosZy .node .label,#mermaid-svg-jV25FQZBShFgosZy .image-shape .label,#mermaid-svg-jV25FQZBShFgosZy .icon-shape .label{text-align:center;}#mermaid-svg-jV25FQZBShFgosZy .node.clickable{cursor:pointer;}#mermaid-svg-jV25FQZBShFgosZy .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-jV25FQZBShFgosZy .arrowheadPath{fill:#333333;}#mermaid-svg-jV25FQZBShFgosZy .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-jV25FQZBShFgosZy .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-jV25FQZBShFgosZy .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-jV25FQZBShFgosZy .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-jV25FQZBShFgosZy .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-jV25FQZBShFgosZy .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-jV25FQZBShFgosZy .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-jV25FQZBShFgosZy .cluster text{fill:#333;}#mermaid-svg-jV25FQZBShFgosZy .cluster span{color:#333;}#mermaid-svg-jV25FQZBShFgosZy div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-jV25FQZBShFgosZy .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-jV25FQZBShFgosZy rect.text{fill:none;stroke-width:0;}#mermaid-svg-jV25FQZBShFgosZy .icon-shape,#mermaid-svg-jV25FQZBShFgosZy .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-jV25FQZBShFgosZy .icon-shape p,#mermaid-svg-jV25FQZBShFgosZy .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-jV25FQZBShFgosZy .icon-shape .label rect,#mermaid-svg-jV25FQZBShFgosZy .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-jV25FQZBShFgosZy .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-jV25FQZBShFgosZy .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-jV25FQZBShFgosZy :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

ETL Pipeline 三大组件

read() / get()

transform() / apply()

write() / accept()

DocumentReaderextendsSupplier<List<Document>>

DocumentTransformerextendsFunction<List<Document>, List<Document>>

DocumentWriterextendsConsumer<List<Document>>

向量数据库

2.1 DocumentReader ——“搬运工”

public interface DocumentReader extends Supplier<List<Document>> {
default List<Document> read() {
return get();
}
}

这个接口极其简单，继承自 Supplier，核心任务就是返回一个 Document 列表。Spring AI 内置了多种实现：

用 JsonReader 举例，它还能用 JSON Pointer 精准提取嵌套数据：

// 基本用法：读取整个 JSON
JsonReader jsonReader = new JsonReader(resource);
List<Document> docs = jsonReader.get();

// 精准提取：只看 items 数组里的内容
List<Document> docs = jsonReader.get("/items");

// 指定字段：只取 description 和 features
JsonReader jsonReader = new JsonReader(resource, "description", "features");

2.2 DocumentTransformer ——“加工车间”

public interface DocumentTransformer extends Function<List<Document>, List<Document>> {
default List<Document> transform(List<Document> documents) {
return apply(documents);
}
}

它继承自 Function，输入一堆 Document，输出加工后的一堆 Document。这是 ETL 中最关键的一环，直接影响 RAG 效果。Spring AI 提供了三大类 Transformer：

（1）TextSplitter 文本分割器

把大文档切成小块。TokenTextSplitter 是最常用的实现：

// 默认参数
TokenTextSplitter splitter = new TokenTextSplitter();

// 自定义参数
TokenTextSplitter splitter = new TokenTextSplitter(
1000, // defaultChunkSize: 每块目标大小（token）
400, // minChunkSizeChars: 每块最小字符数
10, // minChunkLengthToEmbed: 最小嵌入长度
5000, // maxNumChunks: 最多生成块数
true // keepSeparator: 是否保留分隔符
);

它的工作原理很聪明：

#mermaid-svg-2dZYQWA8IB1lRlO8{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-2dZYQWA8IB1lRlO8 .error-icon{fill:#552222;}#mermaid-svg-2dZYQWA8IB1lRlO8 .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-2dZYQWA8IB1lRlO8 .marker{fill:#333333;stroke:#333333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .marker.cross{stroke:#333333;}#mermaid-svg-2dZYQWA8IB1lRlO8 svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-2dZYQWA8IB1lRlO8 p{margin:0;}#mermaid-svg-2dZYQWA8IB1lRlO8 .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster-label text{fill:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster-label span{color:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster-label span p{background-color:transparent;}#mermaid-svg-2dZYQWA8IB1lRlO8 .label text,#mermaid-svg-2dZYQWA8IB1lRlO8 span{fill:#333;color:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .node rect,#mermaid-svg-2dZYQWA8IB1lRlO8 .node circle,#mermaid-svg-2dZYQWA8IB1lRlO8 .node ellipse,#mermaid-svg-2dZYQWA8IB1lRlO8 .node polygon,#mermaid-svg-2dZYQWA8IB1lRlO8 .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .rough-node .label text,#mermaid-svg-2dZYQWA8IB1lRlO8 .node .label text,#mermaid-svg-2dZYQWA8IB1lRlO8 .image-shape .label,#mermaid-svg-2dZYQWA8IB1lRlO8 .icon-shape .label{text-anchor:middle;}#mermaid-svg-2dZYQWA8IB1lRlO8 .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .rough-node .label,#mermaid-svg-2dZYQWA8IB1lRlO8 .node .label,#mermaid-svg-2dZYQWA8IB1lRlO8 .image-shape .label,#mermaid-svg-2dZYQWA8IB1lRlO8 .icon-shape .label{text-align:center;}#mermaid-svg-2dZYQWA8IB1lRlO8 .node.clickable{cursor:pointer;}#mermaid-svg-2dZYQWA8IB1lRlO8 .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .arrowheadPath{fill:#333333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-2dZYQWA8IB1lRlO8 .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-2dZYQWA8IB1lRlO8 .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-2dZYQWA8IB1lRlO8 .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster text{fill:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 .cluster span{color:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-2dZYQWA8IB1lRlO8 .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-2dZYQWA8IB1lRlO8 rect.text{fill:none;stroke-width:0;}#mermaid-svg-2dZYQWA8IB1lRlO8 .icon-shape,#mermaid-svg-2dZYQWA8IB1lRlO8 .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-2dZYQWA8IB1lRlO8 .icon-shape p,#mermaid-svg-2dZYQWA8IB1lRlO8 .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-2dZYQWA8IB1lRlO8 .icon-shape .label rect,#mermaid-svg-2dZYQWA8IB1lRlO8 .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-2dZYQWA8IB1lRlO8 .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-2dZYQWA8IB1lRlO8 .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-2dZYQWA8IB1lRlO8 :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

输入文本

1. CL100K_BASE 编码成 Token

2. 按 defaultChunkSize 切块

3. 在句号/问号/换行处找断点

4. 保留语义完整的文本块

输出 Document 列表

大白话：它不是死板地每 500 字切一刀，而是切完后会"回头看看"——如果正好切在句子中间，就往前或往后挪到最近的句号处。这样保证每块都是完整的话。

（2）MetadataEnricher 元数据增强器

给文档"贴标签"，方便后续精准检索：

// 自动提取关键词标签
KeywordMetadataEnricher enricher = new KeywordMetadataEnricher(chatModel, 5);
List<Document> enrichedDocs = enricher.apply(documents);

// 自动生成摘要（还能关联前后文档）
SummaryMetadataEnricher enricher = new SummaryMetadataEnricher(
chatModel,
List.of(SummaryType.PREVIOUS, SummaryType.CURRENT, SummaryType.NEXT)
);

大白话：就像图书馆给每本书贴标签（“科技类”“2025年”“AI”），以后找书时先看标签就能快速筛掉不相关的。

（3）ContentFormatter 内容格式化工具

统一文档格式，控制哪些元数据跟着内容一起喂给 AI。通过 DefaultContentFormatter 的 Builder 模式灵活配置：

DefaultContentFormatter formatter = DefaultContentFormatter.builder()
.withMetadataTemplate("{key}: {value}")
.withMetadataSeparator("\\n")
.withTextTemplate("{metadata_string}\\n\\n{content}")
.withExcludedInferenceMetadataKeys("embedding", "vector_id")
.build();

2.3 DocumentWriter ——“入库员”

public interface DocumentWriter extends Consumer<List<Document>> {
default void write(List<Document> documents) {
accept(documents);
}
}

它继承自 Consumer，负责把处理好的文档存到目标位置。Spring AI 提供了两种内置实现：

• FileDocumentWriter：写入文件系统（调试用）
• VectorStore：写入向量数据库（生产用）

VectorStore 支持的向量数据库一览：

#mermaid-svg-ZSMC4f9sHHyf8rbl{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-ZSMC4f9sHHyf8rbl .error-icon{fill:#552222;}#mermaid-svg-ZSMC4f9sHHyf8rbl .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-ZSMC4f9sHHyf8rbl .marker{fill:#333333;stroke:#333333;}#mermaid-svg-ZSMC4f9sHHyf8rbl .marker.cross{stroke:#333333;}#mermaid-svg-ZSMC4f9sHHyf8rbl svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-ZSMC4f9sHHyf8rbl p{margin:0;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge{stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 path{fill:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 text{fill:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon–1{font-size:40px;color:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge–1{stroke:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth–1{stroke-width:17;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section–1 line{stroke:hsl(60, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 path{fill:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-0{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-0{stroke:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-0{stroke-width:14;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-0 line{stroke:hsl(240, 100%, 83.5294117647%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 path{fill:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-1{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-1{stroke:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-1{stroke-width:11;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-1 line{stroke:hsl(260, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 path{fill:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 text{fill:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-2{font-size:40px;color:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-2{stroke:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-2{stroke-width:8;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 line{stroke:hsl(90, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 path{fill:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-3{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-3{stroke:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-3{stroke-width:5;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-3 line{stroke:hsl(120, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 path{fill:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-4{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-4{stroke:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-4{stroke-width:2;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-4 line{stroke:hsl(150, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 path{fill:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-5{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-5{stroke:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-5{stroke-width:-1;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-5 line{stroke:hsl(180, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 path{fill:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-6{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-6{stroke:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-6{stroke-width:-4;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-6 line{stroke:hsl(210, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 path{fill:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-7{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-7{stroke:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-7{stroke-width:-7;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-7 line{stroke:hsl(270, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 path{fill:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-8{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-8{stroke:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-8{stroke-width:-10;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-8 line{stroke:hsl(330, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 path{fill:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-9{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-9{stroke:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-9{stroke-width:-13;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-9 line{stroke:hsl(0, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 polygon,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 path{fill:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 text{fill:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .node-icon-10{font-size:40px;color:black;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-edge-10{stroke:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge-depth-10{stroke-width:-16;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-10 line{stroke:hsl(30, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:lightgray;}#mermaid-svg-ZSMC4f9sHHyf8rbl .disabled text{fill:#efefef;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root rect,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root path,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root circle,#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root polygon{fill:hsl(240, 100%, 46.2745098039%);}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root text{fill:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-root span{color:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .section-2 span{color:#ffffff;}#mermaid-svg-ZSMC4f9sHHyf8rbl .icon-container{height:100%;display:flex;justify-content:center;align-items:center;}#mermaid-svg-ZSMC4f9sHHyf8rbl .edge{fill:none;}#mermaid-svg-ZSMC4f9sHHyf8rbl .mindmap-node-label{dy:1em;alignment-baseline:middle;text-anchor:middle;dominant-baseline:middle;text-align:center;}#mermaid-svg-ZSMC4f9sHHyf8rbl :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

VectorStore

Milvus

Redis

PGVector

Elasticsearch

MongoDB Atlas

Weaviate

Qdrant

Pinecone

阿里云 DashScope

三、完整 ETL 流水线实战

把三大组件串起来，一个完整的 ETL 流程就是：

// 抽取：从 PDF 读取
PDFReader pdfReader = new PagePdfDocumentReader("knowledge_base.pdf");
List<Document> documents = pdfReader.read();

// 转换：分割 + 添加摘要
TokenTextSplitter splitter = new TokenTextSplitter(500, 50);
List<Document> splitDocs = splitter.apply(documents);
SummaryMetadataEnricher enricher = new SummaryMetadataEnricher(chatModel,
List.of(SummaryType.CURRENT));
List<Document> enrichedDocs = enricher.apply(splitDocs);

// 加载：写入向量库
vectorStore.write(enrichedDocs);

// 或者用函数式链式调用，一行搞定！
vectorStore.write(enricher.apply(splitter.apply(pdfReader.read())));

#mermaid-svg-kQD1GwrZMjxNJ9fY{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-kQD1GwrZMjxNJ9fY .error-icon{fill:#552222;}#mermaid-svg-kQD1GwrZMjxNJ9fY .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-kQD1GwrZMjxNJ9fY .marker{fill:#333333;stroke:#333333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .marker.cross{stroke:#333333;}#mermaid-svg-kQD1GwrZMjxNJ9fY svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-kQD1GwrZMjxNJ9fY p{margin:0;}#mermaid-svg-kQD1GwrZMjxNJ9fY .label{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;color:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster-label text{fill:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster-label span{color:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster-label span p{background-color:transparent;}#mermaid-svg-kQD1GwrZMjxNJ9fY .label text,#mermaid-svg-kQD1GwrZMjxNJ9fY span{fill:#333;color:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .node rect,#mermaid-svg-kQD1GwrZMjxNJ9fY .node circle,#mermaid-svg-kQD1GwrZMjxNJ9fY .node ellipse,#mermaid-svg-kQD1GwrZMjxNJ9fY .node polygon,#mermaid-svg-kQD1GwrZMjxNJ9fY .node path{fill:#ECECFF;stroke:#9370DB;stroke-width:1px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .rough-node .label text,#mermaid-svg-kQD1GwrZMjxNJ9fY .node .label text,#mermaid-svg-kQD1GwrZMjxNJ9fY .image-shape .label,#mermaid-svg-kQD1GwrZMjxNJ9fY .icon-shape .label{text-anchor:middle;}#mermaid-svg-kQD1GwrZMjxNJ9fY .node .katex path{fill:#000;stroke:#000;stroke-width:1px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .rough-node .label,#mermaid-svg-kQD1GwrZMjxNJ9fY .node .label,#mermaid-svg-kQD1GwrZMjxNJ9fY .image-shape .label,#mermaid-svg-kQD1GwrZMjxNJ9fY .icon-shape .label{text-align:center;}#mermaid-svg-kQD1GwrZMjxNJ9fY .node.clickable{cursor:pointer;}#mermaid-svg-kQD1GwrZMjxNJ9fY .root .anchor path{fill:#333333!important;stroke-width:0;stroke:#333333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .arrowheadPath{fill:#333333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edgePath .path{stroke:#333333;stroke-width:2.0px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .flowchart-link{stroke:#333333;fill:none;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edgeLabel{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-kQD1GwrZMjxNJ9fY .edgeLabel p{background-color:rgba(232,232,232, 0.8);}#mermaid-svg-kQD1GwrZMjxNJ9fY .edgeLabel rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-kQD1GwrZMjxNJ9fY .labelBkg{background-color:rgba(232, 232, 232, 0.5);}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster rect{fill:#ffffde;stroke:#aaaa33;stroke-width:1px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster text{fill:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY .cluster span{color:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY div.mermaidTooltip{position:absolute;text-align:center;max-width:200px;padding:2px;font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:12px;background:hsl(80, 100%, 96.2745098039%);border:1px solid #aaaa33;border-radius:2px;pointer-events:none;z-index:100;}#mermaid-svg-kQD1GwrZMjxNJ9fY .flowchartTitleText{text-anchor:middle;font-size:18px;fill:#333;}#mermaid-svg-kQD1GwrZMjxNJ9fY rect.text{fill:none;stroke-width:0;}#mermaid-svg-kQD1GwrZMjxNJ9fY .icon-shape,#mermaid-svg-kQD1GwrZMjxNJ9fY .image-shape{background-color:rgba(232,232,232, 0.8);text-align:center;}#mermaid-svg-kQD1GwrZMjxNJ9fY .icon-shape p,#mermaid-svg-kQD1GwrZMjxNJ9fY .image-shape p{background-color:rgba(232,232,232, 0.8);padding:2px;}#mermaid-svg-kQD1GwrZMjxNJ9fY .icon-shape .label rect,#mermaid-svg-kQD1GwrZMjxNJ9fY .image-shape .label rect{opacity:0.5;background-color:rgba(232,232,232, 0.8);fill:rgba(232,232,232, 0.8);}#mermaid-svg-kQD1GwrZMjxNJ9fY .label-icon{display:inline-block;height:1em;overflow:visible;vertical-align:-0.125em;}#mermaid-svg-kQD1GwrZMjxNJ9fY .node .label-icon path{fill:currentColor;stroke:revert;stroke-width:revert;}#mermaid-svg-kQD1GwrZMjxNJ9fY :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

完整 ETL 流程

knowledge_base.pdf

PagePdfDocumentReader按页读取

List<Document>（原始文档）

TokenTextSplitter按 Token 智能切分

List<Document>（分割后）

SummaryMetadataEnricherAI 生成摘要元数据

List<Document>（增强后）

VectorStore.write()向量化 + 入库

向量数据库

四、Spring AI Alibaba 的"外挂"扩展

如果说 Spring AI 提供了 ETL 的"标配"，那 Spring AI Alibaba 社区 就是给它装上了"豪华大礼包"。它扩展了大量的 DocumentReader 和 DocumentParser，几乎覆盖了你日常开发中能遇到的所有数据源。

4.1 DocumentReader 扩展

4.2 DocumentParser 扩展

Parser 与 Reader 的区别：Reader 负责"从哪读"，Parser 负责"怎么解析格式"。

4.3 Reader + Parser 配合使用

Reader 和 Parser 经常搭档出场。比如从语雀读取一篇 PDF 附件：

// Parser 负责解析 PDF 格式
TikaDocumentParser parser = new TikaDocumentParser();

// Reader 负责从语雀获取文档，并用 Parser 解析
YuQueResource resource = YuQueResource.builder()
.yuQueToken("your-token")
.resourcePath("your-path")
.build();
YuQueDocumentReader reader = new YuQueDocumentReader(resource, parser);

// 读取
List<Document> documents = reader.get();

大白话：Reader 是"快递员"（负责从各种地方取件），Parser 是"拆包员"（负责把包裹里的东西解析出来）。有的 Reader 自带拆包能力，有的则需要外挂 Parser。

五、ETL 最佳实践

文档质量决定了 AI 回答能力的上限，其他优化策略只是让 AI 回答能力不断逼近上限。

5.1 文档处理三原则

（1）内容结构化

原始文档排版清晰、标题层次分明、列表不要嵌套太深。建议使用 Markdown 作为知识库文档的首选格式——它天生结构清晰，解析效果好。

（2）内容规范化

• 语言统一：中文场景用中文文档，别中英混杂
• 表述统一：同一概念用同一说法（"机器学习"别一会儿叫 ML，一会儿叫 Machine Learning）
• 减少噪音：去掉水印、复杂表格等干扰元素

（3）格式标准化

优先使用 Markdown、DOCX 等文本格式。PDF 解析效果相对较差，可以先用百炼 DashScopeParse 工具转成 Markdown 再处理。

5.2 文档切片技巧

切片是 ETL 中最考验经验的环节：

最佳策略：智能分块 + 人工二次校验。先用算法自动切分，再人工抽查几篇，确认语义完整性。

// 调整切片参数
TokenTextSplitter splitter = new TokenTextSplitter(
200, // 减小块大小
100, // 降低最小字符数
10, // 最小嵌入长度
5000,
true
);

5.3 元数据标注策略

给你的知识碎片打上丰富的标签，形成多维索引：

// 方式1：手动添加
new Document("案例内容…", Map.of(
"type", "interior",
"year", "2025",
"style", "modern"
));

// 方式2：AI 自动提取关键词
KeywordMetadataEnricher enricher = new KeywordMetadataEnricher(chatModel, 5);
List<Document> enriched = enricher.apply(documents);

// 方式3：批量为文档添加来源标签
MarkdownDocumentReaderConfig config = MarkdownDocumentReaderConfig.builder()
.withAdditionalMetadata("source", "codefather.cn")
.withAdditionalMetadata("status", "已审核")
.build();

5.4 批处理策略

当文档量很大时，别一次性全塞给嵌入模型——可能触发 API 限制。Spring AI 提供了 BatchingStrategy 来智能分批：

// 自定义分批策略
@Bean
public BatchingStrategy customTokenCountBatchingStrategy() {
return new TokenCountBatchingStrategy(
EncodingType.CL100K_BASE, // 编码类型
8000, // 最大 token 数
0.1 // 保留百分比（余量）
);
}

大白话：就像搬家时不能把整个房间的东西塞进一个箱子，得分批打包。每批不能超过搬运工（嵌入模型）的承重上限。

六、总结

#mermaid-svg-8AhEiZqUthZGUkkJ{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;fill:#333;}@keyframes edge-animation-frame{from{stroke-dashoffset:0;}}@keyframes dash{to{stroke-dashoffset:0;}}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-animation-slow{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 50s linear infinite;stroke-linecap:round;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-animation-fast{stroke-dasharray:9,5!important;stroke-dashoffset:900;animation:dash 20s linear infinite;stroke-linecap:round;}#mermaid-svg-8AhEiZqUthZGUkkJ .error-icon{fill:#552222;}#mermaid-svg-8AhEiZqUthZGUkkJ .error-text{fill:#552222;stroke:#552222;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-thickness-normal{stroke-width:1px;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-thickness-thick{stroke-width:3.5px;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-pattern-solid{stroke-dasharray:0;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-thickness-invisible{stroke-width:0;fill:none;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-pattern-dashed{stroke-dasharray:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-pattern-dotted{stroke-dasharray:2;}#mermaid-svg-8AhEiZqUthZGUkkJ .marker{fill:#333333;stroke:#333333;}#mermaid-svg-8AhEiZqUthZGUkkJ .marker.cross{stroke:#333333;}#mermaid-svg-8AhEiZqUthZGUkkJ svg{font-family:\”trebuchet ms\”,verdana,arial,sans-serif;font-size:16px;}#mermaid-svg-8AhEiZqUthZGUkkJ p{margin:0;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge{stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 path{fill:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 text{fill:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon–1{font-size:40px;color:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge–1{stroke:hsl(240, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth–1{stroke-width:17;}#mermaid-svg-8AhEiZqUthZGUkkJ .section–1 line{stroke:hsl(60, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 path{fill:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-0{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-0{stroke:hsl(60, 100%, 73.5294117647%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-0{stroke-width:14;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-0 line{stroke:hsl(240, 100%, 83.5294117647%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 path{fill:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-1{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-1{stroke:hsl(80, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-1{stroke-width:11;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-1 line{stroke:hsl(260, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 path{fill:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 text{fill:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-2{font-size:40px;color:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-2{stroke:hsl(270, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-2{stroke-width:8;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 line{stroke:hsl(90, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 path{fill:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-3{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-3{stroke:hsl(300, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-3{stroke-width:5;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-3 line{stroke:hsl(120, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 path{fill:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-4{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-4{stroke:hsl(330, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-4{stroke-width:2;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-4 line{stroke:hsl(150, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 path{fill:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-5{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-5{stroke:hsl(0, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-5{stroke-width:-1;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-5 line{stroke:hsl(180, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 path{fill:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-6{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-6{stroke:hsl(30, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-6{stroke-width:-4;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-6 line{stroke:hsl(210, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 path{fill:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-7{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-7{stroke:hsl(90, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-7{stroke-width:-7;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-7 line{stroke:hsl(270, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 path{fill:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-8{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-8{stroke:hsl(150, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-8{stroke-width:-10;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-8 line{stroke:hsl(330, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 path{fill:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-9{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-9{stroke:hsl(180, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-9{stroke-width:-13;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-9 line{stroke:hsl(0, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 polygon,#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 path{fill:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 text{fill:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .node-icon-10{font-size:40px;color:black;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-edge-10{stroke:hsl(210, 100%, 76.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .edge-depth-10{stroke-width:-16;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-10 line{stroke:hsl(30, 100%, 86.2745098039%);stroke-width:3;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled circle,#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:lightgray;}#mermaid-svg-8AhEiZqUthZGUkkJ .disabled text{fill:#efefef;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-root rect,#mermaid-svg-8AhEiZqUthZGUkkJ .section-root path,#mermaid-svg-8AhEiZqUthZGUkkJ .section-root circle,#mermaid-svg-8AhEiZqUthZGUkkJ .section-root polygon{fill:hsl(240, 100%, 46.2745098039%);}#mermaid-svg-8AhEiZqUthZGUkkJ .section-root text{fill:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-root span{color:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .section-2 span{color:#ffffff;}#mermaid-svg-8AhEiZqUthZGUkkJ .icon-container{height:100%;display:flex;justify-content:center;align-items:center;}#mermaid-svg-8AhEiZqUthZGUkkJ .edge{fill:none;}#mermaid-svg-8AhEiZqUthZGUkkJ .mindmap-node-label{dy:1em;alignment-baseline:middle;text-anchor:middle;dominant-baseline:middle;text-align:center;}#mermaid-svg-8AhEiZqUthZGUkkJ :root{–mermaid-font-family:\”trebuchet ms\”,verdana,arial,sans-serif;}

Spring AI ETL

Extract 抽取

DocumentReader

JsonReader

TextReader

MarkdownReader

PDFReader

HtmlReader

TikaDocumentReader

Alibaba 扩展

POI / GitHub / Notion

飞书 / 语雀 / B站

MySQL / MongoDB / ES

Transform 转换

TextSplitter

TokenTextSplitter

MetadataEnricher

KeywordMetadataEnricher

SummaryMetadataEnricher

ContentFormatter

DefaultContentFormatter

Load 加载

DocumentWriter

FileDocumentWriter

VectorStore

PGVector 实战

阿里云 PostgreSQL

pgvector 插件

这篇文章我们系统地学习了 Spring AI 的 ETL Pipeline：

掌握了 ETL，你就掌握了 RAG 系统的"任督二脉"。接下来就可以搭建自己的知识库，让 AI 真正"读懂"你的专属数据了。

参考文档：

• Spring AI 官方文档 – ETL Pipeline
• Spring AI Alibaba – ETL Pipeline
• Spring AI Alibaba – Document Reader 使用指南
• Spring AI Alibaba – Document Parser 使用指南