Merge pull request 'fix(avatar): index knowledge documents asynchronously' (#15) from codex/avatar-knowledge-async-20260904 into main

Reviewed-on: #15

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

@@ -17,8 +17,7 @@ UPLOAD_DIR = os.path.abspath(os.getenv("UPLOAD_DIR", os.path.join(BASE_DIR, "upl
 os.makedirs(UPLOAD_DIR, exist_ok=True)
 
 ALLOWED_EXT = {".md", ".txt", ".pdf", ".doc", ".docx", ".xlsx"}
-MAX_UPLOAD_BYTES = 50 * 1024 * 1024
+MAX_UPLOAD_BYTES = 10 * 1024 * 1024
-UPLOAD_CHUNK_BYTES = 1024 * 1024
 
 
 class QAIn(BaseModel):
@@ -81,25 +80,17 @@ async def upload_doc(avatar_id: str, file: UploadFile = File(...), authorization
     os.makedirs(avatar_dir, exist_ok=True)
     stored = f"{uuid.uuid4().hex}{ext}"
     path = os.path.join(avatar_dir, stored)
-    file_size = 0
+    content = await file.read()
-    try:
+    if len(content) > MAX_UPLOAD_BYTES:
-        # Stream large files to disk so a 100MB upload does not occupy 100MB RAM.
+        return fail("文件不能超过 10MB", code=400)
-        with open(path, "wb") as f:
+    with open(path, "wb") as f:
-            while chunk := await file.read(UPLOAD_CHUNK_BYTES):
+        f.write(content)
-                file_size += len(chunk)
-                if file_size > MAX_UPLOAD_BYTES:
-                    raise ValueError("文件不能超过 50MB")
-                f.write(chunk)
-    except ValueError as exc:
-        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_size=len(content),
         file_url=f"/api/files/{avatar_id}/{stored}",
         status="parsing",
     )

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

@@ -64,29 +64,6 @@ def test_upload_returns_before_background_vectorization(
         db.close()
 
 
-def test_upload_rejects_oversize_file_before_queuing_indexing(
-    tmp_path: Path,
-    authorization_context,
-):
-    context = authorization_context
-    with (
-        patch("routers.knowledge.UPLOAD_DIR", str(tmp_path)),
-        patch("routers.knowledge.MAX_UPLOAD_BYTES", 4),
-        patch("routers.knowledge.knowledge_vectorizer.enqueue") as enqueue,
-    ):
-        response = client.post(
-            f"/api/avatar/{context['avatar'].id}/knowledge/docs",
-            headers=context["owner_headers"],
-            files={"file": ("oversize.md", b"12345", "text/markdown")},
-        )
-
-    payload = response.json()
-    assert payload["code"] == 400
-    assert payload["message"] == "文件不能超过 50MB"
-    enqueue.assert_not_called()
-    assert not list((tmp_path / context["avatar"].id).glob("*"))
-
-
 def test_background_vectorizer_commits_ready_document_and_chunks_together(
     tmp_path: Path,
     authorization_context,

digital-avatar-app/docs/H5_PRODUCTION_DEPLOYMENT.md

@@ -117,7 +117,7 @@ location /api/ {
     proxy_set_header X-Forwarded-Proto $scheme;
     proxy_buffering off;
     proxy_read_timeout 300s;
-    client_max_body_size 100m;
+    client_max_body_size 20m;
 }
 ```
 

digital-avatar-app/nginx.conf

@@ -23,10 +23,6 @@ http {
 
         root /usr/share/nginx/html;
         index index.html;
-        # Keep the application gateway aligned with the production edge gateway.
-        # Without this Nginx rejects ordinary PDF uploads with HTTP 413 before
-        # FastAPI can return its user-facing file-size validation message.
-        client_max_body_size 100m;
 
         # SPA 兜底（hash 路由下深链接也可正常加载）
         location / {