fix: harden rag streaming and queue status reporting

docs/RAG/RAG聊天接口.md

@@ -1,6 +1,7 @@
 # RAG 聊天接口
 
-> 最后整理：2026-05-07
+> 最后整理：2026-05-11
+> 本文已按当前后端实现逐项核对
 > 对应后端：`fastapi_modules/fastapi_leaudit/controllers/ragChatController.py`
 > 统一前缀：`/api/v3/rag`
 
@@ -237,6 +238,27 @@ JWT payload 至少会被后端消费这些字段：
 }
 ```
 
+补充说明：
+
+- 如果 `appId` 不存在、不可见，或者系统默认应用 / 排序第一条应用均不可见，当前实现不会报 404，而是返回一个空参数对象：
+
+```json
+{
+  "code": 200,
+  "msg": "success",
+  "data": {
+    "openingStatement": "",
+    "suggestedQuestions": [],
+    "userInputForm": [],
+    "fileUpload": {
+      "image": {
+        "enabled": false
+      }
+    }
+  }
+}
+```
+
 ### 4.5 发起流式对话
 
 `POST /api/v3/rag/chat/messages`
@@ -256,7 +278,7 @@ JWT payload 至少会被后端消费这些字段：
 | 字段 | 类型 | 必填 | 说明 |
 |------|------|------|------|
 | `query` | string | 是 | 用户问题，不能为空 |
-| `conversationId` | string \| null | 否 | 会话 ID；新对话可传 `null` 或不传 |
+| `conversationId` | string \| null | 否 | 会话 ID；新对话可传 `null`、不传，当前实现也把 `"-1"` 视为新对话 |
 | `appId` | int \| null | 否 | 应用 ID；不传则自动回退默认应用 |
 
 返回类型：
@@ -291,6 +313,7 @@ data: {"event":"error","task_id":"...","message_id":"...","code":"llm_error","me
 - 流结束后会落一条 `role = assistant` 消息
 - 若命中知识库，会把引用结果写入 `sources / metadata`
 - 会根据对话内容追加 `suggested_questions`
+- 当前应用解析顺序是：指定 `appId` -> 任意默认应用 -> 排序第一条应用；每一步都只检查当前命中的那一条记录是否可见，不会遍历全部可见应用
 
 ### 4.6 获取会话列表
 
@@ -382,6 +405,7 @@ data: {"event":"error","task_id":"...","message_id":"...","code":"llm_error","me
 说明：
 
 - 返回结构是按“问答对”聚合后的结果，不是底层 `rag_message` 原始逐条结果。
+- 分页查询实际先按底层 `rag_message` 行分页，再在服务层折叠成“用户问 + 助手答”结构，所以单页返回条数可能少于 `pageSize`。
 
 ### 4.8 重命名会话
 
@@ -440,12 +464,16 @@ data: {"event":"error","task_id":"...","message_id":"...","code":"llm_error","me
 }
 ```
 
-可选值：
+约定值：
 
 - `like`
 - `dislike`
 - `null`
 
+补充说明：
+
+- 当前 DTO 仅声明 `rating: str | null`，后端不会强校验枚举，除 `null` 外其余字符串也会被写入 `rag_message.feedback`。
+
 成功响应示例：
 
 ```json
@@ -474,6 +502,7 @@ data: {"event":"error","task_id":"...","message_id":"...","code":"llm_error","me
 - `fileUpload.image.enabled` 固定为 `false`
 - 检索依赖 Chroma；Chroma 不可用时，接口仍可回答，但会退化成无知识库上下文
 - 建议问题 `suggestedQuestions` 由二次模型调用生成，失败时会降级为空数组
+- `GET /chat/parameters` / `POST /chat/messages` 的应用回退逻辑目前不是“从全部可见应用里挑第一条”，而是“默认应用优先，其次全表第一条”；当默认应用或首条应用不可见时，可能出现空参数或 `未配置可用聊天应用`
 
 ## 7. 联调建议
 

fastapi_modules/fastapi_leaudit/controllers/documentController.py

@@ -294,64 +294,62 @@ class DocumentController(BaseController):
         async def GetQueueStatus():
             """获取文档处理队列状态。"""
             from datetime import datetime
+            from fastapi_admin.config import LEAUDIT_WORKER_CONCURRENCY
 
             async with GetAsyncSession() as Session:
-                statusRows = (
+                runStats = (
                     await Session.execute(
                         text(
                             """
-                            SELECT processing_status, COUNT(*) AS cnt
+                            SELECT
-                            FROM leaudit_documents
+                                COUNT(*) FILTER (WHERE r.status IN ('queued', 'pending', 'retrying'))::int AS pending_tasks,
-                            WHERE deleted_at IS NULL AND is_latest_version = true
+                                COUNT(*) FILTER (WHERE r.status = 'running')::int AS processing_tasks
-                            GROUP BY processing_status
+                            FROM leaudit_audit_runs r
+                            JOIN leaudit_documents d
+                              ON d.current_run_id = r.id
+                            WHERE d.deleted_at IS NULL
+                              AND d.is_latest_version = true
                             """
                         )
                     )
-                ).mappings().all()
+                ).mappings().first()
 
-            waiting = 0
+                pending = int((runStats or {}).get("pending_tasks") or 0)
-            processing = 0
+                processing = int((runStats or {}).get("processing_tasks") or 0)
-            for row in statusRows:
-                s = str(row["processing_status"] or "")
-                c = int(row["cnt"] or 0)
-                if s == "waiting":
-                    waiting = c
-                elif s in ("processing", "running"):
-                    processing += c
 
-            processingIdsRows: list[int] = []
+                processingIdsRows = (
-            if processing > 0:
+                    await Session.execute(
-                async with GetAsyncSession() as Session:
+                        text(
-                    idRows = (
+                            """
-                        await Session.execute(
+                            SELECT d.id
-                            text(
+                            FROM leaudit_documents d
-                                """
+                            JOIN leaudit_audit_runs r
-                                SELECT id FROM leaudit_documents
+                              ON r.id = d.current_run_id
-                                WHERE deleted_at IS NULL
+                            WHERE d.deleted_at IS NULL
-                                  AND is_latest_version = true
+                              AND d.is_latest_version = true
-                                  AND processing_status IN ('processing', 'running')
+                              AND r.status = 'running'
-                                ORDER BY updated_at DESC
+                            ORDER BY COALESCE(r.started_at, r.created_at) DESC, d.id DESC
-                                LIMIT 50
+                            LIMIT 50
-                                """
+                            """
-                            )
                         )
-                    ).fetchall()
+                    )
-                    processingIdsRows = [int(r[0]) for r in idRows]
+                ).fetchall()
+                processingIds = [int(r[0]) for r in processingIdsRows]
 
             return Result.success(
                 data=QueueStatusVO(
                     success=True,
                     timestamp=datetime.now().isoformat(),
                     queue={
-                        "pending_tasks": waiting,
+                        "pending_tasks": pending,
                         "processing_tasks": processing,
-                        "available_slots": max(0, 4 - processing),
+                        "available_slots": max(0, LEAUDIT_WORKER_CONCURRENCY - processing),
-                        "max_concurrent": 4,
+                        "max_concurrent": LEAUDIT_WORKER_CONCURRENCY,
                     },
                     documents={
-                        "waiting": waiting,
+                        "waiting": pending,
                         "processing": processing,
-                        "processing_ids": processingIdsRows,
+                        "processing_ids": processingIds,
                     },
                 )
             )

fastapi_modules/fastapi_leaudit/rag_engine/generator.py

@@ -77,7 +77,9 @@ async def generate_stream(
                     if payload == "[DONE]":
                         break
                     chunk = json.loads(payload)
-                    delta = chunk.get("choices", [{}])[0].get("delta", {})
+                    choices = chunk.get("choices") or []
+                    first_choice = choices[0] if choices and isinstance(choices[0], dict) else {}
+                    delta = first_choice.get("delta", {})
                     text = delta.get("content", "")
                     if text:
                         yield _sse_line(
@@ -141,4 +143,4 @@ async def generate_stream(
 
 
 def _sse_line(data: dict) -> str:
-    return f"data: {json.dumps(data, ensure_ascii=False)}\\n\\n"
+    return f"data: {json.dumps(data, ensure_ascii=False)}\n\n"