[前端+RAG] 恢复同步上传修复导读生成；用Uint8Array存PDF字节修复detached；CSS覆盖PDF阅读模式空白

chat_web_front/src/views/reading/index.vue

@@ -283,7 +283,14 @@ provide('selectedFile', selectedFile);
 const docHtml = ref('');
 const fileContent = ref(null);
 const readingBox = ref(null);
-const pdfData = ref<ArrayBuffer | null>(null);
+const pdfBytes = ref<Uint8Array | null>(null);  // 存原始字节，不会被 detach
+const pdfData = computed(() => {
+  // 每次访问时复制一份新的 ArrayBuffer 给 PdfViewer
+  if (!pdfBytes.value) return null;
+  const copy = new ArrayBuffer(pdfBytes.value.byteLength);
+  new Uint8Array(copy).set(pdfBytes.value);
+  return copy;
+});
 const readingMode = ref(false);
 const fileType = computed(() => {
   const name = selectedFile.value?.fileName || '';
@@ -447,7 +454,7 @@ const handleNodeClick = async (data: any) => {
   if (ext === 'pdf') {
     await loadPdfFile();
   } else {
-    pdfData.value = null;
+    pdfBytes.value = null;
     await loadFileContent();
   }
 };
@@ -460,13 +467,9 @@ const loadPdfFile = async () => {
       params: { fileId: selectedFile.value.fileId },
       responseType: 'arraybuffer'
     });
-    // 复制 ArrayBuffer 避免被 Vue 响应式代理导致 detached
+    pdfBytes.value = new Uint8Array(resp.data as ArrayBuffer);
-    const src = resp.data as ArrayBuffer;
-    const copy = new ArrayBuffer(src.byteLength);
-    new Uint8Array(copy).set(new Uint8Array(src));
-    pdfData.value = copy;
   } catch (e: any) {
-    pdfData.value = null;
+    pdfBytes.value = null;
     docHtml.value = '<p style="color:#999;text-align:center;margin-top:40px;">PDF 文件加载失败</p>';
   }
   // 同时加载 HTML 用于笔记功能（后台）
@@ -911,6 +914,9 @@ onMounted(async () => {
       flex: 1; overflow: auto; position: relative; padding: 0;
       .view-md {
         padding: 20px;
+        // 覆盖 PyMuPDF get_text("html") 输出的固定宽度
+        :deep(div[style*="width:"]) { width: 100% !important; max-width: 100% !important; }
+        :deep(.pdf-page > div) { width: 100% !important; }
         :deep(p) { font-size: 15px; line-height: 1.8rem; margin-block-start: 0; }
         :deep(.highlight) { background: #D0EAC8; }
         :deep(.note-flag) { width: 23px; height: 28px; line-height: 28px; display: inline-block; text-align: center; font-weight: bold; font-size: 10px; margin-left: 8px; cursor: pointer; background: url("@/assets/images/reading/note.png"); color: #004EA0; background-size: contain !important; background-repeat: no-repeat !important; background-position: center bottom !important; }

langchain-chat/server/knowledge_base/kb_doc_api.py

@@ -269,62 +269,6 @@ def upload_docs(
     return BaseResponse(code=200, msg="文件上传与向量化完成", data={"failed_files": failed_files})
 
 
-def _background_llm_and_vectorize(
-    knowledge_base_name: str,
-    file_names: List[str],
-    chunk_size: int,
-    chunk_overlap: int,
-    zh_title_enhance: bool,
-    docs: dict,
-    not_refresh_vs_cache: bool,
-):
-    """后台线程：执行 LLM 导读生成 + 向量化，不阻塞上传响应。"""
-    import time
-    start_time = time.time()
-
-    kb = KBServiceFactory.get_service_by_name(knowledge_base_name)
-
-    # 1. 生成 LLM 导读（摘要、关键词、章节速览）
-    for filename in file_names:
-        try:
-            knowledge_file = KnowledgeFile(filename=filename, knowledge_base_name=knowledge_base_name)
-            new_loop = asyncio.new_event_loop()
-            asyncio.set_event_loop(new_loop)
-            try:
-                llm_result = new_loop.run_until_complete(knowledge_file.get_llm_result())
-            finally:
-                new_loop.close()
-
-            # 将 LLM 结果写入缓存文件，供 Java 后端轮询读取
-            import json
-            cache_dir = os.path.join(os.path.dirname(os.path.dirname(os.path.dirname(__file__))), "knowledge_base", knowledge_base_name)
-            os.makedirs(cache_dir, exist_ok=True)
-            cache_file = os.path.join(cache_dir, f"{filename}.llm_result.json")
-            with open(cache_file, 'w', encoding='utf-8') as f:
-                json.dump(llm_result, f, ensure_ascii=False)
-            logger.info(f"[后台] LLM 导读生成完成: {filename}")
-        except Exception as e:
-            logger.error(f"[后台] LLM 导读生成失败 {filename}: {e}")
-
-    # 2. 向量化
-    try:
-        _update_docs_impl(
-            knowledge_base_name=knowledge_base_name,
-            file_names=file_names,
-            override_custom_docs=True,
-            chunk_size=chunk_size,
-            chunk_overlap=chunk_overlap,
-            zh_title_enhance=zh_title_enhance,
-            docs=docs,
-            not_refresh_vs_cache=True,
-        )
-        if kb and not not_refresh_vs_cache:
-            kb.save_vector_store()
-        logger.info(f"[后台] 向量化完成，总耗时: {time.time() - start_time:.2f}s")
-    except Exception as e:
-        logger.error(f"[后台] 向量化失败: {e}")
-
-
 def upload_docs_new(
         files: List[UploadFile] = File(..., description="上传文件，支持多文件"),
         knowledge_base_name: str = Form(..., description="知识库名称", examples=["samples"]),
@@ -338,7 +282,7 @@ def upload_docs_new(
         not_refresh_vs_cache: bool = Form(False, description="暂不保存向量库（用于FAISS）"),
 ) -> BaseResponse:
     """
-    API接口：上传文件，先提取全文快速返回，LLM导读+向量化后台异步执行
+    API接口：上传文件，并/或向量化
     """
     import time
     start_time = time.time()
@@ -360,50 +304,63 @@ def upload_docs_new(
     file_names = list(docs.keys())
     llm_results = {}
 
-    # 保存文件到磁盘 + 提取全文（快速操作）
+    # 先将上传的文件保存到磁盘
     for result in _save_files_in_thread(files, knowledge_base_name=knowledge_base_name, override=override):
         filename = result["data"]["file_name"]
         if result["code"] != 200:
             failed_files[filename] = result["msg"]
+
         if filename not in file_names:
             file_names.append(filename)
 
-        # 仅提取全文（快速），不调用 LLM
+        # 生成摘要、关键词、章节速览（模型已优化为 deepseek-v3）
         try:
             knowledge_file = KnowledgeFile(filename=filename, knowledge_base_name=knowledge_base_name)
-            full_text_data = knowledge_file.get_full_text()
+            import concurrent.futures
-            import json as _json
+            def run_async_in_thread():
-            try:
+                new_loop = asyncio.new_event_loop()
-                full_text = _json.loads(full_text_data).get("full_text", "")
+                asyncio.set_event_loop(new_loop)
-            except:
+                try:
-                full_text = ""
+                    return new_loop.run_until_complete(knowledge_file.get_llm_result())
+                finally:
+                    new_loop.close()
+
+            with concurrent.futures.ThreadPoolExecutor(max_workers=1) as executor:
+                future = executor.submit(run_async_in_thread)
+                llm_result = future.result()
             llm_results[filename] = {
-                "full_text": full_text,
+                "full_text": llm_result.get("full_text", "获取全文失败"),
-                "article_abstract": "导读生成中...",
+                "article_abstract": llm_result.get("article_abstract", "生成摘要失败"),
-                "article_keywords": "导读生成中...",
+                "article_keywords": llm_result.get("article_keywords", "生成关键词失败"),
-                "article_paragraph": "导读生成中..."
+                "article_paragraph": llm_result.get("article_paragraph", "生成章节速览失败")
             }
         except Exception as e:
-            logger.error(f"提取全文失败 {filename}: {e}")
+            logger.error(f"生成LLM结果时出错：{e}", exc_info=e if log_verbose else None)
             llm_results[filename] = {
-                "full_text": "",
+                "article_abstract": "生成摘要失败",
-                "article_abstract": "导读生成中...",
+                "article_keywords": "生成关键词失败",
-                "article_keywords": "导读生成中...",
+                "article_paragraph": "生成章节速览失败"
-                "article_paragraph": "导读生成中..."
             }
 
-    # 后台异步执行 LLM 导读 + 向量化（不阻塞响应）
+    # 对保存的文件进行向量化
-    import threading
+    if to_vector_store:
-    bg_thread = threading.Thread(
+        update_st = time.time()
-        target=_background_llm_and_vectorize,
+        result = _update_docs_impl(
-        args=(knowledge_base_name, file_names, chunk_size, chunk_overlap,
+            knowledge_base_name=knowledge_base_name,
-              zh_title_enhance, docs, not_refresh_vs_cache),
+            file_names=file_names,
-        daemon=True
+            override_custom_docs=True,
-    )
+            chunk_size=chunk_size,
-    bg_thread.start()
+            chunk_overlap=chunk_overlap,
-
+            zh_title_enhance=zh_title_enhance,
-    logger.info(f"文件上传+全文提取用时: {time.time() - start_time:.2f}s，LLM+向量化已转后台")
+            docs=docs,
-    return BaseResponse(code=200, msg="文件上传完成，导读生成中", data={
+            not_refresh_vs_cache=True,
+        )
+        failed_files.update(result.data["failed_files"])
+        if not not_refresh_vs_cache:
+            kb.save_vector_store()
+        logger.info(f'向量化用时：{time.time() - update_st}')
+    logger.info(f"总执行时间: {time.time() - start_time:.2f}s")
+    return BaseResponse(code=200, msg="文件上传与向量化完成", data={
         "failed_files": failed_files,
         "llm_results": llm_results
     })