fix(avatar): upload knowledge files in chunks

Reviewed-on: #17

digital-avatar-app/backend/database.py

@@ -54,6 +54,8 @@ def init_db():
         ("knowledge_docs", "chunk_count", "INTEGER DEFAULT 0"),
         ("knowledge_docs", "vectorized_at", "TIMESTAMP"),
         ("knowledge_docs", "error_message", "VARCHAR DEFAULT ''"),
+        ("knowledge_docs", "index_stage", "VARCHAR DEFAULT ''"),
+        ("knowledge_docs", "index_progress", "INTEGER DEFAULT 0"),
         ("avatars", "owner_id", "VARCHAR DEFAULT ''"),
         ("authorizations", "takeover_enabled", "BOOLEAN DEFAULT 0"),
         ("authorizations", "takeover_mode", "VARCHAR DEFAULT 'immediate'"),

digital-avatar-app/backend/embeddings.py

@@ -51,7 +51,7 @@ def _hash_embedding(texts, dim=EMBED_DIM):
     return vecs
 
 
-def embed(texts):
+def embed(texts, on_progress=None):
     """返回 list[list[float]]，与输入顺序一致。"""
     if not texts:
         return []
@@ -64,6 +64,7 @@ def embed(texts):
         except ValueError:
             batch_size = 10
         embeddings = []
+        total = len(texts)
         for start in range(0, len(texts), batch_size):
             batch = texts[start:start + batch_size]
             payload = json.dumps({"input": batch, "model": model}).encode("utf-8")
@@ -84,8 +85,13 @@ def embed(texts):
             if len(items) != len(batch):
                 raise ValueError("embedding response count does not match request")
             embeddings.extend(item["embedding"] for item in items)
+            if on_progress:
+                on_progress(len(embeddings), total)
         return embeddings
-    return _hash_embedding(texts)
+    vectors = _hash_embedding(texts)
+    if on_progress:
+        on_progress(len(vectors), len(texts))
+    return vectors
 
 
 def cosine(a, b):

digital-avatar-app/backend/models.py

@@ -191,6 +191,8 @@ class KnowledgeDoc(Base):
     file_url = Column(String, default="")
     status = Column(String, default="uploaded")  # uploaded | parsing | ready | failed
     error_message = Column(String, default="")  # 建立索引失败原因
+    index_stage = Column(String, default="")  # queued | extracting | chunking | embedding | ready | failed
+    index_progress = Column(Integer, default=0)  # 0-100
     vectorized = Column(Boolean, default=False)  # 是否已向量化
     embedding_model = Column(String, default="")  # 向量模型标识
     chunk_count = Column(Integer, default=0)  # 切片数量
@@ -207,6 +209,8 @@ class KnowledgeDoc(Base):
             "fileUrl": self.file_url,
             "status": self.status,
             "errorMessage": self.error_message or "",
+            "indexStage": self.index_stage or "",
+            "indexProgress": int(self.index_progress or 0),
             "vectorized": bool(self.vectorized),
             "embeddingModel": self.embedding_model,
             "chunkCount": self.chunk_count,

digital-avatar-app/backend/routers/knowledge.py

@@ -1,4 +1,7 @@
 import os
+import json
+import shutil
+import time
 import uuid
 
 from fastapi import APIRouter, UploadFile, File, Depends, Header, HTTPException
@@ -19,6 +22,9 @@ os.makedirs(UPLOAD_DIR, exist_ok=True)
 ALLOWED_EXT = {".md", ".txt", ".pdf", ".doc", ".docx", ".xlsx"}
 MAX_UPLOAD_BYTES = 50 * 1024 * 1024
 UPLOAD_CHUNK_BYTES = 1024 * 1024
+MULTIPART_CHUNK_BYTES = 5 * 1024 * 1024
+MULTIPART_ROOT = ".multipart"
+MULTIPART_TTL_SECONDS = 24 * 60 * 60
 
 
 class QAIn(BaseModel):
@@ -31,6 +37,74 @@ class EnabledIn(BaseModel):
     enabled: bool = True
 
 
+class MultipartUploadIn(BaseModel):
+    filename: str
+    fileSize: int
+    totalChunks: int
+
+
+def _validate_document(filename: str, file_size: int):
+    ext = os.path.splitext(filename or "")[1].lower()
+    if ext not in ALLOWED_EXT:
+        return None, f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx"
+    if file_size <= 0:
+        return None, "文件内容不能为空"
+    if file_size > MAX_UPLOAD_BYTES:
+        return None, "文件不能超过 50MB"
+    return ext, ""
+
+
+def _multipart_dir(avatar_id: str, upload_id: str) -> str:
+    safe_avatar_id = os.path.basename(avatar_id)
+    safe_upload_id = os.path.basename(upload_id)
+    if (
+        safe_avatar_id != avatar_id
+        or safe_upload_id != upload_id
+        or len(upload_id) != 32
+        or any(character not in "0123456789abcdef" for character in upload_id)
+    ):
+        raise HTTPException(status_code=400, detail="上传标识无效")
+    return os.path.join(UPLOAD_DIR, MULTIPART_ROOT, safe_avatar_id, safe_upload_id)
+
+
+def _purge_stale_multipart_uploads(avatar_id: str):
+    avatar_upload_root = os.path.join(UPLOAD_DIR, MULTIPART_ROOT, os.path.basename(avatar_id))
+    if not os.path.isdir(avatar_upload_root):
+        return
+    cutoff = time.time() - MULTIPART_TTL_SECONDS
+    for entry in os.scandir(avatar_upload_root):
+        if entry.is_dir(follow_symlinks=False) and entry.stat(follow_symlinks=False).st_mtime < cutoff:
+            shutil.rmtree(entry.path, ignore_errors=True)
+
+
+def _read_multipart_metadata(avatar_id: str, upload_id: str) -> tuple[str, dict]:
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    metadata_path = os.path.join(upload_dir, "metadata.json")
+    if not os.path.isfile(metadata_path):
+        raise HTTPException(status_code=404, detail="上传任务不存在或已过期")
+    with open(metadata_path, "r", encoding="utf-8") as stream:
+        return upload_dir, json.load(stream)
+
+
+def _create_knowledge_doc(db: Session, avatar_id: str, filename: str, ext: str, file_size: int, stored: str):
+    doc = KnowledgeDoc(
+        id=uuid.uuid4().hex,
+        avatar_id=avatar_id,
+        filename=filename,
+        file_type=ext.lstrip("."),
+        file_size=file_size,
+        file_url=f"/api/files/{avatar_id}/{stored}",
+        status="parsing",
+        index_stage="queued",
+        index_progress=0,
+    )
+    db.add(doc)
+    db.commit()
+    db.refresh(doc)
+    knowledge_vectorizer.enqueue(doc.id)
+    return doc
+
+
 def _doc_payload(doc: KnowledgeDoc) -> dict:
     payload = doc.to_dict()
     stored_name = os.path.basename(doc.file_url or "")
@@ -74,9 +148,9 @@ def list_docs(avatar_id: str, authorization: str = Header(None), db: Session = D
 @router.post("/avatar/{avatar_id}/knowledge/docs")
 async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization: str = Header(None), db: Session = Depends(get_db)):
     _require_owned_avatar(db, avatar_id, authorization)
-    ext = os.path.splitext(file.filename or "")[1].lower()
-    if ext not in ALLOWED_EXT:
-        return fail(f"不支持的文件类型：{ext or '空'}，仅支持 md/txt/pdf/doc/docx/xlsx", code=400)
+    ext, validation_error = _validate_document(file.filename or "", 1)
+    if validation_error:
+        return fail(validation_error, code=400)
     avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
     os.makedirs(avatar_dir, exist_ok=True)
     stored = f"{uuid.uuid4().hex}{ext}"
@@ -94,23 +168,124 @@ async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization
         if os.path.exists(path):
             os.remove(path)
         return fail(str(exc), code=400)
-    doc = KnowledgeDoc(
-        id=uuid.uuid4().hex,
-        avatar_id=avatar_id,
-        filename=file.filename,
-        file_type=ext.lstrip("."),
-        file_size=file_size,
-        file_url=f"/api/files/{avatar_id}/{stored}",
-        status="parsing",
+    if file_size == 0:
+        if os.path.exists(path):
+            os.remove(path)
+        return fail("文件内容不能为空", code=400)
+
+    doc = _create_knowledge_doc(db, avatar_id, file.filename or stored, ext, file_size, stored)
+    return ok(_doc_payload(doc))
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads")
+def create_multipart_upload(
+    avatar_id: str,
+    body: MultipartUploadIn,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    ext, validation_error = _validate_document(body.filename, body.fileSize)
+    if validation_error:
+        return fail(validation_error, code=400)
+    expected_chunks = (body.fileSize + MULTIPART_CHUNK_BYTES - 1) // MULTIPART_CHUNK_BYTES
+    if body.totalChunks != expected_chunks:
+        return fail("文件分片数量不正确", code=400)
+
+    _purge_stale_multipart_uploads(avatar_id)
+    upload_id = uuid.uuid4().hex
+    upload_dir = _multipart_dir(avatar_id, upload_id)
+    os.makedirs(upload_dir, exist_ok=False)
+    metadata = {
+        "filename": body.filename,
+        "fileSize": body.fileSize,
+        "totalChunks": body.totalChunks,
+        "extension": ext,
+    }
+    with open(os.path.join(upload_dir, "metadata.json"), "w", encoding="utf-8") as stream:
+        json.dump(metadata, stream, ensure_ascii=False)
+    return ok({"uploadId": upload_id, "chunkSize": MULTIPART_CHUNK_BYTES})
+
+
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/chunks/{chunk_index}")
+async def upload_multipart_chunk(
+    avatar_id: str,
+    upload_id: str,
+    chunk_index: int,
+    file: UploadFile = File(...),
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    if chunk_index < 0 or chunk_index >= total_chunks:
+        return fail("文件分片序号不正确", code=400)
+
+    expected_size = min(
+        MULTIPART_CHUNK_BYTES,
+        int(metadata["fileSize"]) - chunk_index * MULTIPART_CHUNK_BYTES,
     )
+    part_path = os.path.join(upload_dir, f"{chunk_index}.part")
+    temporary_path = f"{part_path}.uploading"
+    received = 0
+    try:
+        with open(temporary_path, "wb") as stream:
+            while chunk := await file.read(UPLOAD_CHUNK_BYTES):
+                received += len(chunk)
+                if received > expected_size:
+                    raise ValueError("文件分片大小不正确")
+                stream.write(chunk)
+        if received != expected_size:
+            raise ValueError("文件分片大小不正确")
+        os.replace(temporary_path, part_path)
+    except ValueError as exc:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        return fail(str(exc), code=400)
+    return ok({"chunkIndex": chunk_index, "uploadedBytes": received})
 
-    # Persist and acknowledge the upload first. Extraction and embeddings may take
-    # minutes for a PDF and must never consume the browser request timeout.
-    db.add(doc)
-    db.commit()
-    db.refresh(doc)
-    knowledge_vectorizer.enqueue(doc.id)
 
+@router.post("/avatar/{avatar_id}/knowledge/uploads/{upload_id}/complete")
+def complete_multipart_upload(
+    avatar_id: str,
+    upload_id: str,
+    authorization: str = Header(None),
+    db: Session = Depends(get_db),
+):
+    _require_owned_avatar(db, avatar_id, authorization)
+    upload_dir, metadata = _read_multipart_metadata(avatar_id, upload_id)
+    total_chunks = int(metadata["totalChunks"])
+    part_paths = [os.path.join(upload_dir, f"{index}.part") for index in range(total_chunks)]
+    if not all(os.path.isfile(path) for path in part_paths):
+        return fail("文件分片尚未上传完整", code=400)
+    if sum(os.path.getsize(path) for path in part_paths) != int(metadata["fileSize"]):
+        return fail("文件分片总大小不正确", code=400)
+
+    avatar_dir = os.path.join(UPLOAD_DIR, avatar_id)
+    os.makedirs(avatar_dir, exist_ok=True)
+    stored = f"{uuid.uuid4().hex}{metadata['extension']}"
+    final_path = os.path.join(avatar_dir, stored)
+    temporary_path = f"{final_path}.assembling"
+    try:
+        with open(temporary_path, "wb") as output:
+            for part_path in part_paths:
+                with open(part_path, "rb") as source:
+                    shutil.copyfileobj(source, output, UPLOAD_CHUNK_BYTES)
+        os.replace(temporary_path, final_path)
+        doc = _create_knowledge_doc(
+            db,
+            avatar_id,
+            metadata["filename"],
+            metadata["extension"],
+            int(metadata["fileSize"]),
+            stored,
+        )
+    except Exception:
+        if os.path.exists(temporary_path):
+            os.remove(temporary_path)
+        raise
+    shutil.rmtree(upload_dir, ignore_errors=True)
     return ok(_doc_payload(doc))
 
 
@@ -134,6 +309,8 @@ def retry_doc(avatar_id: str, doc_id: str, authorization: str = Header(None), db
     doc.chunk_count = 0
     doc.vectorized_at = None
     doc.error_message = ""
+    doc.index_stage = "queued"
+    doc.index_progress = 0
     db.commit()
     db.refresh(doc)
     knowledge_vectorizer.enqueue(doc.id)

digital-avatar-app/backend/services/knowledge_vectorizer.py

@@ -74,11 +74,19 @@ class KnowledgeVectorizer:
             if not stored_name or not os.path.isfile(path):
                 raise FileNotFoundError("原文件不可用，请重新上传")
 
+            self._set_progress(db, doc, "extracting", 8)
             text = embeddings.extract_text(path, f".{doc.file_type}")
+            self._set_progress(db, doc, "chunking", 22)
             chunks = embeddings.chunk_text(text)
             if not chunks:
                 raise ValueError("文档没有可建立索引的文字内容")
-            vectors = embeddings.embed(chunks)
+            self._set_progress(db, doc, "embedding", 30)
+
+            def embedding_progress(done: int, total: int):
+                percent = 30 + int((done / max(1, total)) * 65)
+                self._set_progress(db, doc, "embedding", min(percent, 95))
+
+            vectors = embeddings.embed(chunks, on_progress=embedding_progress)
             if len(vectors) != len(chunks):
                 raise ValueError("向量服务返回数量与文档分段不一致")
 
@@ -103,6 +111,8 @@ class KnowledgeVectorizer:
             doc.vectorized_at = datetime.now(timezone.utc)
             doc.status = "ready"
             doc.error_message = ""
+            doc.index_stage = "ready"
+            doc.index_progress = 100
             db.commit()
             logger.info("Knowledge document %s indexed with %s chunks", doc.id, len(chunks))
         except Exception as exc:
@@ -116,10 +126,18 @@ class KnowledgeVectorizer:
                 failed_doc.chunk_count = 0
                 failed_doc.vectorized_at = None
                 failed_doc.error_message = str(exc)[:300] or "建立知识索引失败"
+                failed_doc.index_stage = "failed"
+                failed_doc.index_progress = 0
                 db.commit()
             logger.exception("Knowledge vectorization failed for %s: %s", doc_id, exc)
         finally:
             db.close()
 
+    @staticmethod
+    def _set_progress(db, doc, stage: str, progress: int):
+        doc.index_stage = stage
+        doc.index_progress = progress
+        db.commit()
+
 
 knowledge_vectorizer = KnowledgeVectorizer()

digital-avatar-app/backend/tests/test_embeddings.py

@@ -49,6 +49,7 @@ class RemoteEmbeddingTests(unittest.TestCase):
         texts = [f"chunk-{index}" for index in range(14)]
         batch_sizes = []
         requested_urls = []
+        progress_updates = []
 
         def fake_urlopen(request, timeout):
             self.assertEqual(timeout, 30)
@@ -68,7 +69,10 @@ class RemoteEmbeddingTests(unittest.TestCase):
             "EMBEDDING_MODEL": "text-embedding-v4",
             "EMBEDDING_BATCH_SIZE": "10",
         }), patch("embeddings.urllib.request.urlopen", side_effect=fake_urlopen):
-            result = embeddings.embed(texts)
+            result = embeddings.embed(
+                texts,
+                on_progress=lambda completed, total: progress_updates.append((completed, total)),
+            )
 
         self.assertEqual(batch_sizes, [10, 4])
         self.assertEqual(requested_urls, [
@@ -76,6 +80,7 @@ class RemoteEmbeddingTests(unittest.TestCase):
             "https://embedding.example/v1/embeddings",
         ])
         self.assertEqual(result, [[float(index)] for index in range(14)])
+        self.assertEqual(progress_updates, [(10, 14), (14, 14)])
 
     def test_full_embedding_endpoint_is_not_modified(self):
         self.assertEqual(

digital-avatar-app/backend/tests/test_knowledge_storage.py

@@ -87,6 +87,84 @@ def test_upload_rejects_oversize_file_before_queuing_indexing(
     assert not list((tmp_path / context["avatar"].id).glob("*"))
 
 
+def test_multipart_upload_reassembles_file_before_queuing_indexing(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    content = b"0123456789"
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": len(content), "totalChunks": 3},
+        ).json()["data"]
+
+        for index, chunk in enumerate((content[:4], content[4:8], content[8:])):
+            response = client.post(
+                f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/{index}",
+                headers=context["owner_headers"],
+                files={"file": (f"chunk-{index}", chunk, "application/octet-stream")},
+            )
+            assert response.json()["code"] == 200
+
+        completed = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        ).json()["data"]
+
+    assert completed["status"] == "parsing"
+    assert completed["fileSize"] == len(content)
+    enqueue.assert_called_once_with(completed["id"])
+    stored_path = tmp_path / avatar_id / Path(completed["fileUrl"]).name
+    assert stored_path.read_bytes() == content
+    assert not (tmp_path / ".multipart" / avatar_id / created["uploadId"]).exists()
+
+    db = SessionLocal()
+    try:
+        stored = db.query(KnowledgeDoc).filter(KnowledgeDoc.id == completed["id"]).one()
+        db.delete(stored)
+        db.commit()
+    finally:
+        db.close()
+
+
+def test_multipart_upload_rejects_incomplete_parts(
+    tmp_path: Path,
+    authorization_context,
+):
+    context = authorization_context
+    avatar_id = context["avatar"].id
+    with (
+        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
+        patch("routers.knowledge.MULTIPART_CHUNK_BYTES", 4),
+        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
+    ):
+        created = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads",
+            headers=context["owner_headers"],
+            json={"filename": "large.pdf", "fileSize": 6, "totalChunks": 2},
+        ).json()["data"]
+        client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/chunks/0",
+            headers=context["owner_headers"],
+            files={"file": ("chunk-0", b"0123", "application/octet-stream")},
+        )
+        response = client.post(
+            f"/api/avatar/{avatar_id}/knowledge/uploads/{created['uploadId']}/complete",
+            headers=context["owner_headers"],
+        )
+
+    assert response.json()["code"] == 400
+    assert response.json()["message"] == "文件分片尚未上传完整"
+    enqueue.assert_not_called()
+
+
 def test_background_vectorizer_commits_ready_document_and_chunks_together(
     tmp_path: Path,
     authorization_context,
@@ -116,6 +194,8 @@ def test_background_vectorizer_commits_ready_document_and_chunks_together(
         assert stored.status == "ready"
         assert stored.vectorized is True
         assert stored.chunk_count == 1
+        assert stored.index_stage == "ready"
+        assert stored.index_progress == 100
         assert db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).count() == 1
         db.query(KnowledgeChunk).filter(KnowledgeChunk.doc_id == stored.id).delete()
         db.delete(stored)

digital-avatar-app/src/api/index.ts

@@ -306,6 +306,8 @@ export interface KnowledgeDoc {
   embeddingModel?: string
   chunkCount?: number
   errorMessage?: string
+  indexStage?: string
+  indexProgress?: number
   createdAt: string
 }
 
@@ -331,13 +333,83 @@ export interface SearchResult {
 export const getKnowledgeDocs = (avatarId: string) =>
   request.get<KnowledgeDoc[]>(`/avatar/${avatarId}/knowledge/docs`)
 
-// 上传文档（支持 md/txt/pdf/doc/docx/xlsx）
-export const uploadKnowledgeDoc = (avatarId: string, file: File) => {
+const KNOWLEDGE_UPLOAD_CHUNK_SIZE = 5 * 1024 * 1024
+
+const uploadKnowledgeChunk = async (
+  avatarId: string,
+  uploadId: string,
+  chunkIndex: number,
+  chunk: Blob,
+  onProgress?: (loaded: number) => void
+) => {
+  const form = new FormData()
+  form.append('file', chunk, `chunk-${chunkIndex}`)
+  let reportedLoaded = 0
+  for (let attempt = 1; attempt <= 3; attempt += 1) {
+    try {
+      await request.post(
+        `/avatar/${avatarId}/knowledge/uploads/${uploadId}/chunks/${chunkIndex}`,
+        form,
+        {
+          headers: { 'Content-Type': 'multipart/form-data' },
+          timeout: 2 * 60 * 1000,
+          onUploadProgress: (event) => {
+            reportedLoaded = Math.max(reportedLoaded, Math.min(event.loaded, chunk.size))
+            onProgress?.(reportedLoaded)
+          }
+        }
+      )
+      return
+    } catch (error: any) {
+      const status = Number(error?.response?.status || 0)
+      const retryable = !status || status === 408 || status === 429 || status >= 500
+      if (!retryable || attempt === 3) throw error
+      await new Promise((resolve) => window.setTimeout(resolve, attempt * 800))
+    }
+  }
+}
+
+// 大文件拆成 5MB 分片，避免生产代理的请求体限制拦截整个文件。
+export const uploadKnowledgeDoc = async (
+  avatarId: string,
+  file: File,
+  onUploadProgress?: (loaded: number, total: number) => void
+) => {
+  if (file.size > KNOWLEDGE_UPLOAD_CHUNK_SIZE) {
+    const totalChunks = Math.ceil(file.size / KNOWLEDGE_UPLOAD_CHUNK_SIZE)
+    const upload: any = await request.post(`/avatar/${avatarId}/knowledge/uploads`, {
+      filename: file.name,
+      fileSize: file.size,
+      totalChunks
+    })
+    let uploadedBytes = 0
+    for (let index = 0; index < totalChunks; index += 1) {
+      const start = index * KNOWLEDGE_UPLOAD_CHUNK_SIZE
+      const chunk = file.slice(start, Math.min(start + KNOWLEDGE_UPLOAD_CHUNK_SIZE, file.size))
+      await uploadKnowledgeChunk(
+        avatarId,
+        upload.uploadId,
+        index,
+        chunk,
+        (chunkLoaded) => onUploadProgress?.(uploadedBytes + chunkLoaded, file.size)
+      )
+      uploadedBytes += chunk.size
+      onUploadProgress?.(uploadedBytes, file.size)
+    }
+    return request.post<KnowledgeDoc>(
+      `/avatar/${avatarId}/knowledge/uploads/${upload.uploadId}/complete`,
+      undefined,
+      { timeout: 2 * 60 * 1000 }
+    )
+  }
+
   const form = new FormData()
   form.append('file', file)
   return request.post<KnowledgeDoc>(`/avatar/${avatarId}/knowledge/docs`, form, {
     headers: { 'Content-Type': 'multipart/form-data' },
-    timeout: 120000
+    // A slow mobile uplink must not be mistaken for a failed upload.
+    timeout: 10 * 60 * 1000,
+    onUploadProgress: (event) => onUploadProgress?.(event.loaded, event.total || file.size)
   })
 }
 

digital-avatar-app/src/views/KnowledgeManage.vue

@@ -15,7 +15,7 @@
 
     <template v-else>
       <div class="tab-switcher" role="tablist" aria-label="知识库类型">
-        <button class="tab-btn" :class="{ active: activeTab === 'docs' }" role="tab" :aria-selected="activeTab === 'docs'" @click="activeTab = 'docs'">文档知识库 <b>{{ docs.length }}</b></button>
+        <button class="tab-btn" :class="{ active: activeTab === 'docs' }" role="tab" :aria-selected="activeTab === 'docs'" @click="activeTab = 'docs'">文档知识库 <b>{{ displayDocs.length }}</b></button>
         <button class="tab-btn" :class="{ active: activeTab === 'qa' }" role="tab" :aria-selected="activeTab === 'qa'" @click="activeTab = 'qa'">标准问答对 <b>{{ qaPairs.length }}</b></button>
       </div>
 
@@ -25,14 +25,14 @@
             <div class="upload-icon">📥</div>
             <p class="upload-title"><span class="upload-link">点击上传</span></p>
             <p class="upload-hint">支持 MD / TXT / PDF / DOC / DOCX / XLSX，上传后自动向量化</p>
-            <input ref="fileInput" type="file" accept=".md,.txt,.pdf,.doc,.docx,.xlsx" class="hidden-input" @change="onFileChange" />
+            <input ref="fileInput" type="file" multiple accept=".md,.txt,.pdf,.doc,.docx,.xlsx" class="hidden-input" @change="onFileChange" />
           </div>
-          <p v-if="uploading" class="uploading-text">文件上传中…</p>
+          <p v-if="uploading" class="uploading-text">{{ pendingUploads.length }} 个文件正在上传</p>
           <p v-if="uploadError" class="error-text">{{ uploadError }}</p>
         </div>
 
-        <div v-if="docs.length" class="mobile-card-list">
-          <article v-for="doc in docs" :key="doc.id" class="knowledge-card">
+        <div v-if="displayDocs.length" class="mobile-card-list">
+          <article v-for="doc in displayDocs" :key="doc.id" class="knowledge-card">
             <div class="card-icon">{{ fileEmoji(doc.fileType) }}</div>
             <div class="card-content">
               <div class="card-title-row">
@@ -41,10 +41,13 @@
               </div>
               <p class="card-meta">{{ doc.fileType.toUpperCase() }} · {{ formatSize(doc.fileSize) }} · {{ formatDate(doc.createdAt) }}</p>
               <p class="card-detail">{{ documentState(doc).detail }}</p>
+              <div v-if="documentState(doc).progress !== undefined" class="progress-track" :aria-label="`${documentState(doc).label} ${documentState(doc).progress}%`">
+                <span class="progress-fill" :style="{ width: `${documentState(doc).progress}%` }"></span>
+              </div>
             </div>
             <div class="card-actions">
               <button v-if="documentState(doc).tone === 'failed'" class="card-retry" @click="retryDoc(doc.id)">重新索引</button>
-              <button class="card-delete" @click="removeDoc(doc.id)">删除</button>
+              <button v-if="!doc.localUploading" class="card-delete" @click="removeDoc(doc.id)">{{ doc.localOnly ? '移除' : '删除' }}</button>
             </div>
           </article>
         </div>
@@ -106,8 +109,9 @@ const avatarId = computed(() => pickScopedAvatarId(route.params.avatarId, store.
 const activeTab = ref<'docs' | 'qa'>('docs')
 
 const docs = ref<any[]>([])
+const pendingUploads = ref<any[]>([])
 const qaPairs = ref<any[]>([])
-const uploading = ref(false)
+const uploading = computed(() => pendingUploads.value.some((doc) => doc.localUploading))
 const uploadError = ref('')
 const dragOver = ref(false)
 const fileInput = ref<HTMLInputElement | null>(null)
@@ -118,7 +122,15 @@ const searching = ref(false)
 const searched = ref(false)
 const searchResults = ref<any[]>([])
 
+const displayDocs = computed(() => [...pendingUploads.value, ...docs.value])
+
 const documentState = (doc: any) => {
+  if (doc.localUploading) {
+    return { tone: 'pending', label: '上传中', detail: `正在上传 ${doc.uploadProgress || 0}%`, progress: doc.uploadProgress || 0 }
+  }
+  if (doc.localOnly) {
+    return { tone: 'failed', label: '上传失败', detail: doc.errorMessage || '文件未上传成功，请移除后重试' }
+  }
   if (doc.filePresent === false) {
     return { tone: 'missing', label: '文件缺失', detail: '原文件不可用，请删除后重新上传' }
   }
@@ -126,7 +138,12 @@ const documentState = (doc: any) => {
     return { tone: 'ready', label: '已入库', detail: `已切分 ${doc.chunkCount || 0} 段，可用于对话` }
   }
   if (['uploaded', 'parsing'].includes(String(doc.status || '').toLowerCase())) {
-    return { tone: 'pending', label: '处理中', detail: '正在解析并建立知识索引' }
+    const stage = String(doc.indexStage || 'queued').toLowerCase()
+    const labels: Record<string, string> = {
+      queued: '等待处理', extracting: '解析文档', chunking: '切分文本', embedding: '向量化中'
+    }
+    const progress = Math.max(0, Math.min(99, Number(doc.indexProgress || 0)))
+    return { tone: 'pending', label: labels[stage] || '处理中', detail: `${labels[stage] || '正在建立知识索引'} ${progress}%`, progress }
   }
   return { tone: 'failed', label: '处理失败', detail: doc.errorMessage || '未能建立知识索引，请重新索引或重新上传' }
 }
@@ -174,36 +191,62 @@ const loadQA = async () => {
 const triggerFile = () => fileInput.value?.click()
 
 const onFileChange = (e: Event) => {
-  const f = (e.target as HTMLInputElement).files?.[0]
-  if (f) doUpload(f)
+  const files = Array.from((e.target as HTMLInputElement).files || [])
+  if (files.length) uploadFiles(files)
   ;(e.target as HTMLInputElement).value = ''
 }
 
 const onDrop = (e: DragEvent) => {
   dragOver.value = false
-  const f = e.dataTransfer?.files?.[0]
-  if (f) doUpload(f)
+  const files = Array.from(e.dataTransfer?.files || [])
+  if (files.length) uploadFiles(files)
 }
 
-const doUpload = async (file: File) => {
+const uploadFiles = (files: File[]) => {
   uploadError.value = ''
-  const ext = '.' + (file.name.split('.').pop() || '').toLowerCase()
-  if (!['.md', '.txt', '.pdf', '.doc', '.docx', '.xlsx'].includes(ext)) {
-    uploadError.value = `不支持的类型：${ext}，仅支持 md/txt/pdf/doc/docx/xlsx`
-    return
-  }
   if (!avatarId.value) {
     uploadError.value = '请先创建数字分身'
     return
   }
-  uploading.value = true
+  for (const file of files) {
+    const ext = '.' + (file.name.split('.').pop() || '').toLowerCase()
+    if (!['.md', '.txt', '.pdf', '.doc', '.docx', '.xlsx'].includes(ext)) {
+      uploadError.value = `不支持的类型：${ext}，仅支持 md/txt/pdf/doc/docx/xlsx`
+      continue
+    }
+    void uploadOne(file, ext)
+  }
+}
+
+const uploadOne = async (file: File, ext: string) => {
+  if (!avatarId.value) return
+  const localId = `upload-${Date.now()}-${Math.random().toString(16).slice(2)}`
+  const card = {
+    id: localId,
+    filename: file.name,
+    fileType: ext.slice(1),
+    fileSize: file.size,
+    createdAt: new Date().toISOString(),
+    localUploading: true,
+    localOnly: true,
+    uploadProgress: 0,
+    errorMessage: ''
+  }
+  pendingUploads.value.unshift(card)
   try {
-    await uploadKnowledgeDoc(avatarId.value, file)
-    await loadDocs()
+    const created: any = await uploadKnowledgeDoc(avatarId.value, file, (loaded, total) => {
+      const current = pendingUploads.value.find((doc) => doc.id === localId)
+      if (current) current.uploadProgress = Math.min(99, Math.round((loaded / Math.max(1, total)) * 100))
+    })
+    pendingUploads.value = pendingUploads.value.filter((doc) => doc.id !== localId)
+    docs.value = [created, ...docs.value.filter((doc) => doc.id !== created.id)]
+    startDocumentPolling()
   } catch (e: any) {
-    uploadError.value = e?.message || '上传失败'
-  } finally {
-    uploading.value = false
+    const current = pendingUploads.value.find((doc) => doc.id === localId)
+    if (current) {
+      current.localUploading = false
+      current.errorMessage = e?.message || '上传失败'
+    }
   }
 }
 
@@ -219,6 +262,11 @@ const retryDoc = async (id: string) => {
 }
 
 const removeDoc = async (id: string) => {
+  const local = pendingUploads.value.find((doc) => doc.id === id)
+  if (local?.localOnly) {
+    pendingUploads.value = pendingUploads.value.filter((doc) => doc.id !== id)
+    return
+  }
   if (!avatarId.value) return
   await deleteKnowledgeDoc(avatarId.value, id)
   await loadDocs()
@@ -338,6 +386,8 @@ onUnmounted(stopDocumentPolling)
 .status-pill.missing { color: #B91C1C; background: #FEF2F2; }
 .status-pill.failed { color: #B91C1C; background: #FEF2F2; }
 .card-meta, .card-detail { margin: 5px 0 0; color: #9398AE; font-size: 11px; line-height: 1.4; }.card-detail { color: #8B6B58; }
+.progress-track { width: 100%; height: 4px; margin-top: 8px; overflow: hidden; border-radius: 999px; background: #FDE7D1; }
+.progress-fill { display: block; height: 100%; border-radius: inherit; background: linear-gradient(90deg, #FB923C, #F97316); transition: width .25s ease; }
 .card-actions { flex: 0 0 auto; display: flex; flex-direction: column; align-items: stretch; gap: 6px; }
 .card-delete, .card-retry { align-self: center; border: 0; border-radius: 8px; padding: 7px 9px; font-size: 12px; cursor: pointer; white-space: nowrap; }
 .card-delete { color: #EF4444; background: #FEF2F2; }