feat：LLM 分析文件并生成结构化建议

app/datafiles/storage.py

@@ -2,6 +2,7 @@
 
 目录结构是 DATA_FILE_DIR/<file_id>/，里面有原文件 source 和每张表的 1.parquet、2.parquet……
 DuckDB 只能读本地文件，所以文件不进 MongoDB；单实例部署，本地磁盘就够了
+会话内的表名 t1、t2……由每个会话一个的计数器分配，文件删除、过期后表名也不再使用
 """
 
 from datetime import UTC, datetime
@@ -9,16 +10,16 @@ from pathlib import Path
 from typing import Any, Protocol
 
 from bson.codec_options import CodecOptions
-from pymongo import ASCENDING
+from pymongo import ASCENDING, ReturnDocument
 
-from app.schemas.data_file import DataFileRecord, FileInsights
+from app.schemas.data_file import DataFileRecord, DataTable, FileInsights
 
 # 原文件不带扩展名：calamine 见到 .xls 等扩展名会按扩展名打开，用户的扩展名和内容不符时会失败
 SOURCE_NAME = "source"
 
 
-def file_dir(root: Path, file_id: str) -> Path:
-    return root / file_id
+def table_path(root: Path, file_id: str, table: DataTable) -> Path:
+    return root / file_id / table.parquet
 
 
 def _now() -> datetime:
@@ -34,10 +35,12 @@ class DataFileStore(Protocol):
         """未过期且未删除的记录"""
         ...
 
-    async def find(
-        self, conversation_id: str, *, include_deleted: bool = False
-    ) -> list[DataFileRecord]:
-        """会话内未过期的记录，按上传时间排列"""
+    async def find(self, conversation_id: str) -> list[DataFileRecord]:
+        """会话内未过期且未删除的记录，按上传时间排列"""
+        ...
+
+    async def reserve_table_numbers(self, conversation_id: str, count: int) -> int:
+        """为会话预留 count 个连续的表号，返回第一个；表号从 1 开始，不会重复分配"""
         ...
 
     async def mark_deleted(self, file_id: str) -> None: ...
@@ -54,6 +57,8 @@ class DataFileStore(Protocol):
 class InMemoryDataFileStore:
     def __init__(self) -> None:
         self._records: dict[str, DataFileRecord] = {}
+        # 会话 → 已分配的最大表号
+        self._table_numbers: dict[str, int] = {}
 
     def _purge_expired(self) -> None:
         now = _now()
@@ -71,18 +76,20 @@ class InMemoryDataFileStore:
         record = self._records.get(file_id)
         return None if record is None or record.deleted else record
 
-    async def find(
-        self, conversation_id: str, *, include_deleted: bool = False
-    ) -> list[DataFileRecord]:
+    async def find(self, conversation_id: str) -> list[DataFileRecord]:
         self._purge_expired()
         records = [
             record
             for record in self._records.values()
-            if record.conversation_id == conversation_id
-            and (include_deleted or not record.deleted)
+            if record.conversation_id == conversation_id and not record.deleted
         ]
         return sorted(records, key=lambda record: record.created_at)
 
+    async def reserve_table_numbers(self, conversation_id: str, count: int) -> int:
+        first = self._table_numbers.get(conversation_id, 0) + 1
+        self._table_numbers[conversation_id] = first + count - 1
+        return first
+
     async def mark_deleted(self, file_id: str) -> None:
         if (record := self._records.get(file_id)) is not None:
             self._records[file_id] = record.model_copy(update={"deleted": True})
@@ -101,6 +108,8 @@ class MongoDataFileStore:
         # 读回的时间统一带 UTC 时区；集合名前缀用于测试隔离
         options = CodecOptions(tz_aware=True, tzinfo=UTC)
         self._files = db.get_collection(f"{prefix}data_files", codec_options=options)
+        # 每个会话一条：_id 是会话 ID，tables 是已分配的最大表号；会话一直保留，计数器也不过期
+        self._table_numbers = db.get_collection(f"{prefix}data_file_table_numbers")
 
     async def ensure_indexes(self) -> None:
         # 过期记录由 TTL 索引删除；TTL 每分钟才扫一次，查询时仍按 expires_at 过滤
@@ -118,15 +127,12 @@ class MongoDataFileStore:
         )
         return None if document is None else _record(document)
 
-    async def find(
-        self, conversation_id: str, *, include_deleted: bool = False
-    ) -> list[DataFileRecord]:
-        query: dict[str, Any] = {
+    async def find(self, conversation_id: str) -> list[DataFileRecord]:
+        query = {
             "conversation_id": conversation_id,
+            "deleted": False,
             "expires_at": {"$gt": _now()},
         }
-        if not include_deleted:
-            query["deleted"] = False
         documents = await (
             self._files.find(query)
             .sort([("created_at", ASCENDING), ("_id", ASCENDING)])
@@ -134,6 +140,15 @@ class MongoDataFileStore:
         )
         return [_record(document) for document in documents]
 
+    async def reserve_table_numbers(self, conversation_id: str, count: int) -> int:
+        document = await self._table_numbers.find_one_and_update(
+            {"_id": conversation_id},
+            {"$inc": {"tables": count}},
+            upsert=True,
+            return_document=ReturnDocument.AFTER,
+        )
+        return document["tables"] - count + 1
+
     async def mark_deleted(self, file_id: str) -> None:
         await self._files.update_one({"_id": file_id}, {"$set": {"deleted": True}})