perf: fast-route orchestration and short-reply guard for qingjian

qingjian_cs/app/client.py

@@ -1,12 +1,18 @@
 import asyncio
 import json
+import re
 import time
 from collections import defaultdict
 
 import websockets
 
 from .callbacks import post_tianwang_callback
-from .config import AUTO_QUOTE_WAIT_SECONDS, MESSAGE_DEBOUNCE_SECONDS, QINGJIAN_WS_URI
+from .config import (
+    AUTO_QUOTE_WAIT_SECONDS,
+    MESSAGE_DEBOUNCE_SECONDS,
+    QINGJIAN_WS_URI,
+    SHORT_REPLY_MAX_CHARS,
+)
 from .logger import setup_logger
 from .observability import activity_event, build_trace_id
 from .orchestrator import Orchestrator
@@ -27,7 +33,7 @@ class QingjianClient:
         self.pending_images: dict[str, list[str]] = defaultdict(list)
         self.auto_quote_tasks: dict[str, asyncio.Task] = {}
         self.last_reply_key: dict[str, str] = {}
-        self.recent_outbound: dict[str, tuple[str, float]] = {}
+        self.recent_outbound: list[tuple[str, str, str, float]] = []
 
     @staticmethod
     def _customer_key(data: dict) -> str:
@@ -57,6 +63,7 @@ class QingjianClient:
         text = str(text or "").strip()
         if not text:
             return
+        text = self._shorten_reply(text)
         msg = {
             "msg_id": "",
             "acc_id": data.get("acc_id", ""),
@@ -70,22 +77,47 @@ class QingjianClient:
         }
         activity_event(self.logger, "send_reply_attempt", trace_id=trace_id, customer_id=data.get("from_id", "-"), msg=text)
         await self.send_message(msg)
-        self.recent_outbound[self._customer_key(data)] = (text, time.monotonic())
+        self.recent_outbound.append((str(data.get("acc_id", "")), str(data.get("from_id", "")), text, time.monotonic()))
+        if len(self.recent_outbound) > 200:
+            self.recent_outbound = self.recent_outbound[-200:]
         activity_event(self.logger, "send_reply_success", trace_id=trace_id, customer_id=data.get("from_id", "-"), msg=text)
 
+    @staticmethod
+    def _clean_text(text: str) -> str:
+        t = str(text or "").strip()
+        t = re.sub(r"\s+", "", t)
+        return t
+
+    def _shorten_reply(self, text: str) -> str:
+        max_len = max(8, int(SHORT_REPLY_MAX_CHARS))
+        t = str(text or "").strip()
+        if len(t) <= max_len:
+            return t
+        parts = re.split(r"[。！？!?]", t)
+        head = next((p.strip() for p in parts if p and p.strip()), t)
+        if len(head) > max_len:
+            head = head[:max_len].rstrip("，,;；:： ")
+        return head or t[:max_len]
+
     def _is_outbound_echo(self, data: dict, msg: str) -> bool:
         """
         轻简可能会把我方刚发送文本回推为“收到消息”。
-        对同 customer_key 的“短时间完全相同文本”做回环拦截，避免无限对话。
+        对“短时间完全相同文本”做回环拦截，兼容 acc/from 对调回推，避免无限对话。
         """
-        key = self._customer_key(data)
-        last = self.recent_outbound.get(key)
-        if not last:
+        in_acc = str(data.get("acc_id", ""))
+        in_from = str(data.get("from_id", ""))
+        in_msg = self._clean_text(msg)
+        now = time.monotonic()
+        if not in_msg:
             return False
-        last_msg, ts = last
-        if (time.monotonic() - ts) > 120:
-            return False
-        return str(msg or "").strip() == last_msg
+        for out_acc, out_to, out_msg, ts in reversed(self.recent_outbound):
+            if (now - ts) > 120:
+                break
+            if self._clean_text(out_msg) != in_msg:
+                continue
+            if (out_acc == in_acc and out_to == in_from) or (out_acc == in_from and out_to == in_acc):
+                return True
+        return False
 
     async def _handle_decision(self, data: dict, merged_msg: str, *, auto_quote: bool = False) -> None:
         key = self._customer_key(data)