fix: Switch to download-based Parquet loading to resolve DuckDB httpfs issues in HF Space

backend/app/database.py

@@ -76,32 +76,32 @@ class DatabaseManager:
         is_hf = os.getenv("SPACES_REPO_NAME") is not None or (HF_TOKEN and DATASET_REPO_ID)
         
         if is_hf:
-            logger.info("HF environment detected. Mounting remote Parquet files...")
+            logger.info("HF environment detected. Downloading remote Parquet files...")
             try:
                 from huggingface_hub import list_repo_files, hf_hub_download
                 
-                # 1. 配置 DuckDB HTTP 访问
-                conn.execute("INSTALL httpfs; LOAD httpfs;")
-                if HF_TOKEN:
-                    header_json = f'{{"Authorization": "Bearer {HF_TOKEN}"}}'
-                    conn.execute(f"SET http_headers='{header_json}';")
-                
-                # 2. 动态获取文件列表
+                # 1. 动态获取文件列表
                 all_files = list_repo_files(repo_id=DATASET_REPO_ID, repo_type="dataset")
                 
-                # 3. 股票列表：直接下载到本地 (因为很小且查询频繁)
+                # 2. 股票列表：下载并加载
                 list_file = "data/stock_list.parquet"
                 if list_file in all_files:
                     local_list_path = hf_hub_download(repo_id=DATASET_REPO_ID, filename=list_file, repo_type="dataset")
                     conn.execute(f"CREATE OR REPLACE TABLE stock_list AS SELECT * FROM read_parquet('{local_list_path}')")
                     logger.info("Local stock_list table created from remote parquet")
                 
-                # 4. 挂载日线数据 (保持远程挂载)
-                parquet_files = [f"https://huggingface.co/datasets/{DATASET_REPO_ID}/resolve/main/{f}" 
-                                for f in all_files if f.startswith("data/parquet/") and f.endswith(".parquet")]
+                # 3. 日线数据：下载并加载 (下载模式替代挂载模式)
+                # 注意：如果文件很多，这步可能会花点时间，但比下载整个 .duckdb 快得多
+                parquet_files = [f for f in all_files if f.startswith("data/parquet/") and f.endswith(".parquet")]
                 
                 if parquet_files:
-                    files_sql = ", ".join([f"'{f}'" for f in parquet_files])
+                    local_paths = []
+                    for f in parquet_files:
+                        # HF SDK 会自动处理缓存，已下载的文件不会重复下载
+                        path = hf_hub_download(repo_id=DATASET_REPO_ID, filename=f, repo_type="dataset")
+                        local_paths.append(f"'{path}'")
+                    
+                    files_sql = ", ".join(local_paths)
                     conn.execute(f"CREATE OR REPLACE VIEW stock_daily AS SELECT * FROM read_parquet([{files_sql}])")
                     logger.info(f"Remote stock_daily view created with {len(parquet_files)} partitions")
                 else:
@@ -113,7 +113,7 @@ class DatabaseManager:
                 logger.info(f"Verification successful: {count} stocks found")
                 
             except Exception as e:
-                logger.error(f"Failed to mount remote Parquet: {e}")
+                logger.error(f"Failed to load remote Parquet: {e}")
                 self._create_tables()
         else:
             self._create_tables()