refactor(kb): 文库列表/正文分离 + [[双链]]索引表(可扛大文件)

不再一次拉回整库正文、不再前端扫全文 —— 列表只读元数据,正文按需取,链接走索引。

- store: SaveDoc 维护 size+preview(前 500 字);ListVault 仅 Select 元数据(name/size/preview,
  不含 content);GetDoc 取单篇全文;DocLink 表 + ReplaceDocLinks(入库/编辑时按 from 重建出链)
  + ListLinks。
- gateway: 入库/笔记保存时正则抽 [[链接]]→ReplaceDocLinks 维护索引;
  /kb/vault 改返元数据+预览;新增 /kb/doc(单篇全文) 与 /kb/links(全库双链)。
- 前端:listVault 返元数据,新增 getDoc/listLinks;VaultPanel 列表只展示名/字数,
  选中后 getDoc 按需载正文(带加载态),反链/笔记关系图改用服务端 links 索引(不扫全文)。

验证:curl /kb/vault 仅 name/size/preview;/kb/doc 取单篇;/kb/links 返 3 条双链。
Preview:文库点「架构总览」按需载正文(平台分五层)、反向链接(1)=Dispatcher(来自索引)。tsc+vite+gateway build 通过。

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Blizzard
2026-06-13 16:53:23 +08:00
parent 4fd44380aa
commit 69967ea534
5 changed files with 174 additions and 46 deletions
+23 -2
View File
@@ -123,7 +123,12 @@ export async function createKb(id: Identity, name: string, kind: string): Promis
export interface VaultDoc {
name: string;
content: string;
size?: number;
preview?: string;
}
export interface DocLink {
from: string;
to: string;
}
// ---- 我的 Agent 编排(服务端保存,owner 隔离)----
@@ -169,7 +174,7 @@ export async function listChatModels(): Promise<string[]> {
}
}
// listVault: GET /api/v1/kb/vault —— 某知识库的原始文档(Obsidian 式文库浏览)。
// listVault: GET /api/v1/kb/vault —— 文库列表(仅元数据 + 预览,不拉全文)。
export async function listVault(id: Identity, kb: string): Promise<VaultDoc[]> {
const res = await fetch(`${GATEWAY}/api/v1/kb/vault?kb=${encodeURIComponent(kb)}`, { headers: idHeaders(id) });
const data = (await res.json()) as { docs?: VaultDoc[]; error?: string };
@@ -177,6 +182,22 @@ export async function listVault(id: Identity, kb: string): Promise<VaultDoc[]> {
return data.docs ?? [];
}
// getDoc: GET /api/v1/kb/doc —— 取单篇文档全文(按需加载,避免列表拉全量)。
export async function getDoc(id: Identity, kb: string, name: string): Promise<string> {
const res = await fetch(`${GATEWAY}/api/v1/kb/doc?kb=${encodeURIComponent(kb)}&name=${encodeURIComponent(name)}`, { headers: idHeaders(id) });
const data = (await res.json()) as { content?: string; error?: string };
if (!res.ok) throw new Error(data.error ?? `doc failed: ${res.status}`);
return data.content ?? "";
}
// listLinks: GET /api/v1/kb/links —— 某库全部 [[双链]](反链/笔记关系图,数据小)。
export async function listLinks(id: Identity, kb: string): Promise<DocLink[]> {
const res = await fetch(`${GATEWAY}/api/v1/kb/links?kb=${encodeURIComponent(kb)}`, { headers: idHeaders(id) });
const data = (await res.json()) as { links?: DocLink[]; error?: string };
if (!res.ok) throw new Error(data.error ?? `links failed: ${res.status}`);
return data.links ?? [];
}
// saveNote: POST /api/v1/kb/note —— 新建/编辑笔记(落库 + 按 doc 重入库替换旧块)。
export async function saveNote(id: Identity, kb: string, name: string, content: string): Promise<void> {
const res = await fetch(`${GATEWAY}/api/v1/kb/note`, {
+40 -34
View File
@@ -34,12 +34,15 @@ import {
listKb,
createKb,
listVault,
getDoc,
listLinks,
saveNote,
type IngestEvent,
type KbHit,
type Triple,
type KbInfo,
type VaultDoc,
type DocLink,
type Identity,
} from "../lib/api";
import { GraphView } from "../components/GraphView";
@@ -431,24 +434,15 @@ export function KbView({ identity }: { identity: Identity }) {
);
}
function escapeReg(s: string): string {
return s.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
}
// wikiLinks 从内容中抽出所有 [[名称]](忽略别名部分)。
function wikiLinks(content: string): string[] {
const out: string[] = [];
const re = /\[\[([^\]|]+)(\|[^\]]*)?\]\]/g;
let m: RegExpExecArray | null;
while ((m = re.exec(content)) !== null) out.push(m[1].trim());
return out;
}
// VaultPanelObsidian 式文库 —— 文档列表 / Markdown 阅读+编辑([[双链]]可点)/ 反向链接 / 笔记关系图。
// VaultPanelObsidian 式文库 —— 列表(仅元数据) / 正文按需加载 / [[双链]]可点 / 反链 / 笔记关系图。
// 列表与正文分离 + 链接走服务端索引,不再一次拉回整库正文,可扛十几万字大文件。
function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
const toast = useToast();
const [docs, setDocs] = useState<VaultDoc[]>([]);
const [links, setLinks] = useState<DocLink[]>([]);
const [sel, setSel] = useState<string | null>(null);
const [content, setContent] = useState("");
const [loadingDoc, setLoadingDoc] = useState(false);
const [loading, setLoading] = useState(false);
const [mode, setMode] = useState<"read" | "graph">("read");
const [editing, setEditing] = useState(false);
@@ -460,11 +454,13 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
const load = useCallback(async () => {
setLoading(true);
try {
const d = await listVault(identity, kb);
const [d, lk] = await Promise.all([listVault(identity, kb), listLinks(identity, kb)]);
setDocs(d);
setLinks(lk);
setSel((s) => (d.some((x) => x.name === s) ? s : d[0]?.name ?? null));
} catch {
setDocs([]);
setLinks([]);
} finally {
setLoading(false);
}
@@ -475,6 +471,20 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
setMode("read");
}, [load]);
// 选中文档时按需取全文(不随列表一起拉)。
useEffect(() => {
if (!sel || editing) return;
let alive = true;
setLoadingDoc(true);
getDoc(identity, kb, sel)
.then((c) => alive && setContent(c))
.catch(() => alive && setContent(""))
.finally(() => alive && setLoadingDoc(false));
return () => {
alive = false;
};
}, [sel, kb, identity, editing]);
const names = new Set(docs.map((d) => d.name));
const current = docs.find((d) => d.name === sel);
const open = (name: string) => {
@@ -484,17 +494,9 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
setEditing(false);
}
};
const backlinks = current
? docs.filter((d) => d.name !== current.name && new RegExp(`\\[\\[\\s*${escapeReg(current.name)}(\\|[^\\]]*)?\\s*\\]\\]`).test(d.content))
: [];
// 笔记关系图:文档间 [[双链]] → 三元组(仅保留指向已存在笔记的边)。
const noteTriples: Triple[] = [];
for (const d of docs) {
for (const link of wikiLinks(d.content)) {
if (names.has(link) && link !== d.name) noteTriples.push({ s: d.name, p: "链接", o: link });
}
}
// 反链/笔记关系图来自服务端 [[双链]]索引,不扫全文。
const backlinks = sel ? [...new Set(links.filter((l) => l.to === sel).map((l) => l.from))] : [];
const noteTriples: Triple[] = links.filter((l) => names.has(l.from) && names.has(l.to)).map((l) => ({ s: l.from, p: "链接", o: l.to }));
const startNew = () => {
setCreatingNew(true);
@@ -507,7 +509,7 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
setCreatingNew(false);
setEditing(true);
setDraftName(current.name);
setDraft(current.content);
setDraft(content);
};
const onSave = async () => {
const name = (creatingNew ? draftName : current?.name ?? "").trim();
@@ -521,10 +523,9 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
toast.push("success", `已保存「${name}」(正在重建索引)`);
setEditing(false);
setCreatingNew(false);
// 乐观更新本地 + 选中,再后台刷新。
setDocs((ds) => [...ds.filter((x) => x.name !== name), { name, content: draft }]);
setContent(draft);
setSel(name);
setTimeout(() => void load(), 300);
setTimeout(() => void load(), 400);
} catch (e) {
toast.push("error", (e as Error).message);
} finally {
@@ -609,8 +610,13 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
<h2 className="mb-3 flex items-center gap-2 text-base font-semibold text-slate-100">
<FileText className="h-4 w-4 text-brand-400" />
{current.name}
{current.size != null && <span className="text-[11px] font-normal text-slate-600">{current.size} </span>}
</h2>
<Markdown text={current.content} className="text-sm" onLink={open} />
{loadingDoc ? (
<div className="flex items-center gap-1.5 text-xs text-slate-500"><Loader2 className="h-3.5 w-3.5 animate-spin" /> </div>
) : (
<Markdown text={content} className="text-sm" onLink={open} />
)}
<div className="mt-6 border-t border-line pt-3">
<div className="mb-2 flex items-center gap-1.5 text-xs font-medium text-slate-400">
<Link2 className="h-3.5 w-3.5" /> {backlinks.length}
@@ -620,9 +626,9 @@ function VaultPanel({ identity, kb }: { identity: Identity; kb: string }) {
) : (
<ul className="space-y-1">
{backlinks.map((b) => (
<li key={b.name}>
<button onClick={() => open(b.name)} className="text-xs text-brand-400 hover:underline">
{b.name}
<li key={b}>
<button onClick={() => open(b)} className="text-xs text-brand-400 hover:underline">
{b}
</button>
</li>
))}
+45 -3
View File
@@ -11,6 +11,7 @@ import (
"io"
"net/http"
"path/filepath"
"regexp"
"strings"
"time"
@@ -95,12 +96,28 @@ func (h *Handler) KbSaveNote(c *gin.Context) {
}
owner := userID(c)
_ = h.db.EnsureKB(c.Request.Context(), owner, rawKB(body.KB), "general")
_ = h.db.SaveDoc(c.Request.Context(), owner, rawKB(body.KB), body.Name, body.Content)
// 落库 + 重建索引由后台 runIngest 统一处理(forceDoc=name 保持笔记身份)。
job := newJobID()
go h.runIngest(job, owner, rawKB(body.KB), scopedKB(c, body.KB), body.Name, "", nil, body.Content)
c.JSON(http.StatusAccepted, gin.H{"job_id": job, "name": body.Name})
}
// wikiLinks 从内容抽取所有 [[名称]](忽略别名)去重,用于维护双链索引。
func wikiLinks(s string) []string {
seen := map[string]bool{}
var out []string
for _, m := range wikiRe.FindAllStringSubmatch(s, -1) {
n := strings.TrimSpace(m[1])
if n != "" && !seen[n] {
seen[n] = true
out = append(out, n)
}
}
return out
}
var wikiRe = regexp.MustCompile(`\[\[([^\]|]+)(\|[^\]]*)?\]\]`)
// KbVault: GET /api/v1/kb/vault?kb= —— 某知识库的全部原始文档(名+内容),供 Obsidian 式文库浏览。
func (h *Handler) KbVault(c *gin.Context) {
rows, err := h.db.ListVault(c.Request.Context(), userID(c), rawKB(c.Query("kb")))
@@ -110,11 +127,35 @@ func (h *Handler) KbVault(c *gin.Context) {
}
docs := make([]gin.H, 0, len(rows))
for _, r := range rows {
docs = append(docs, gin.H{"name": r.Name, "content": r.Content})
docs = append(docs, gin.H{"name": r.Name, "size": r.Size, "preview": r.Preview})
}
c.JSON(http.StatusOK, gin.H{"docs": docs})
}
// KbDoc: GET /api/v1/kb/doc?kb=&name= —— 取单篇文档全文(按需加载,不在列表里拉全量)。
func (h *Handler) KbDoc(c *gin.Context) {
d, err := h.db.GetDoc(c.Request.Context(), userID(c), rawKB(c.Query("kb")), c.Query("name"))
if err != nil || d == nil {
c.JSON(http.StatusNotFound, gin.H{"error": "文档不存在"})
return
}
c.JSON(http.StatusOK, gin.H{"name": d.Name, "content": d.Content, "size": d.Size})
}
// KbLinks: GET /api/v1/kb/links?kb= —— 某库全部 [[双链]](from→to),供反链/笔记关系图。
func (h *Handler) KbLinks(c *gin.Context) {
rows, err := h.db.ListLinks(c.Request.Context(), userID(c), rawKB(c.Query("kb")))
if err != nil {
c.JSON(http.StatusBadGateway, gin.H{"error": err.Error()})
return
}
links := make([]gin.H, 0, len(rows))
for _, l := range rows {
links = append(links, gin.H{"from": l.FromName, "to": l.ToName})
}
c.JSON(http.StatusOK, gin.H{"links": links})
}
// noteName 取文本首个非空行作笔记名(截断 40 字),用于文本入库的文库留存。
func noteName(text string) string {
for _, line := range strings.Split(text, "\n") {
@@ -191,7 +232,8 @@ func (h *Handler) runIngest(job, owner, kbName, scoped, forceDoc, filename strin
docName = noteName(text)
}
if text != "" {
_ = h.db.SaveDoc(ctx, owner, kbName, docName, text)
_ = h.db.SaveDoc(ctx, owner, kbName, docName, text, "", len([]rune(text)))
_ = h.db.ReplaceDocLinks(ctx, owner, kbName, docName, wikiLinks(text)) // 维护 [[双链]] 索引
}
// 调 mcp-go kb_ingest(带 job_id):它会发 切块/向量化/写入/完成 事件 + CompleteStream。
+3 -1
View File
@@ -30,7 +30,9 @@ func New(db *store.Postgres, cache *store.Redis, bus *nats.Bus) *gin.Engine {
api.POST("/kb/ingest_file", h.KbIngestFile) // 文件入库(docx/xlsx/pdf… 异步)
api.GET("/kb/ingest/:id/stream", h.KbIngestStream) // 入库进度 SSE(实时监控)
api.POST("/kb/search", h.KbSearch) // 知识库检索台(→ mcp-go kb_search
api.GET("/kb/vault", h.KbVault) // 文库:原始文档浏览(Obsidian 式
api.GET("/kb/vault", h.KbVault) // 文库:文档列表(仅元数据+预览
api.GET("/kb/doc", h.KbDoc) // 取单篇文档全文(按需加载)
api.GET("/kb/links", h.KbLinks) // 某库全部 [[双链]](反链/笔记关系图)
api.POST("/kb/note", h.KbSaveNote) // 新建/编辑笔记(落库 + 按 doc 重入库)
api.GET("/kb/graph", h.KbGraph) // 知识图谱三元组(→ mcp-go kb_graphNeo4j
+63 -6
View File
@@ -111,24 +111,81 @@ type DocLink struct {
func (DocLink) TableName() string { return "sundynix_doc_link" }
// SaveDoc 写入/更新一份文档(owner+kb+name 唯一,重名覆盖内容)
func (p *Postgres) SaveDoc(ctx context.Context, owner, kb, name, content string) error {
// docHead 取前 n 个 rune 作预览
func docHead(s string, n int) string {
r := []rune(s)
if len(r) <= n {
return s
}
return string(r[:n])
}
// SaveDoc 写入/更新一份文档(owner+kb+name 唯一,重名覆盖)。
// 同时维护 size 与 preview(列表只读它们,不拉全文)。content 入参为内联正文;
// objectKey 非空表示正文已转 MinIO(此时 content 传空)。
func (p *Postgres) SaveDoc(ctx context.Context, owner, kb, name, content, objectKey string, size int) error {
if p.db == nil {
return nil
}
preview := docHead(content, 500)
return p.db.WithContext(ctx).Clauses(clause.OnConflict{
Columns: []clause.Column{{Name: "owner"}, {Name: "kb"}, {Name: "name"}},
DoUpdates: clause.AssignmentColumns([]string{"content"}),
}).Create(&Doc{Owner: owner, KB: kb, Name: name, Content: content}).Error
DoUpdates: clause.Assignments(map[string]any{"content": content, "object_key": objectKey, "size": size, "preview": preview, "updated_at": time.Now()}),
}).Create(&Doc{Owner: owner, KB: kb, Name: name, Content: content, ObjectKey: objectKey, Size: size, Preview: preview}).Error
}
// ListVault 返回某 owner 某 kb 的全部文档(名+内容),供文库浏览/双链/反链
// ListVault 返回文库列表(仅元数据 + 预览,不含全文),避免一次拉回整库正文
func (p *Postgres) ListVault(ctx context.Context, owner, kb string) ([]Doc, error) {
if p.db == nil {
return nil, nil
}
var rows []Doc
err := p.db.WithContext(ctx).Where("owner = ? AND kb = ?", owner, kb).Order("id").Find(&rows).Error
err := p.db.WithContext(ctx).
Select("id", "name", "size", "preview", "object_key", "updated_at").
Where("owner = ? AND kb = ?", owner, kb).Order("updated_at desc").Find(&rows).Error
return rows, err
}
// GetDoc 取单篇文档(含全文 Content 与 ObjectKey,供按需阅读)。
func (p *Postgres) GetDoc(ctx context.Context, owner, kb, name string) (*Doc, error) {
if p.db == nil {
return nil, nil
}
var d Doc
if err := p.db.WithContext(ctx).Where("owner = ? AND kb = ? AND name = ?", owner, kb, name).First(&d).Error; err != nil {
return nil, err
}
return &d, nil
}
// ReplaceDocLinks 重建某文档的出链(先删旧,再插新)—— 入库/编辑时调用。
func (p *Postgres) ReplaceDocLinks(ctx context.Context, owner, kb, from string, tos []string) error {
if p.db == nil {
return nil
}
return p.db.WithContext(ctx).Transaction(func(tx *gorm.DB) error {
if err := tx.Where("owner = ? AND kb = ? AND from_name = ?", owner, kb, from).Delete(&DocLink{}).Error; err != nil {
return err
}
for _, to := range tos {
if to == "" || to == from {
continue
}
if err := tx.Create(&DocLink{Owner: owner, KB: kb, FromName: from, ToName: to}).Error; err != nil {
return err
}
}
return nil
})
}
// ListLinks 返回某 kb 的全部 [[双链]](from→to),供反链/笔记关系图按需查询(数据小)。
func (p *Postgres) ListLinks(ctx context.Context, owner, kb string) ([]DocLink, error) {
if p.db == nil {
return nil, nil
}
var rows []DocLink
err := p.db.WithContext(ctx).Where("owner = ? AND kb = ?", owner, kb).Find(&rows).Error
return rows, err
}