fix: return HTTP 400 (not 502) for client ModelError/bad model requests

gateway/scripts/ocg_router.py

@@ -513,6 +513,7 @@ class RouterHandler(BaseHTTPRequestHandler):
 
         last_error = ""
         success = False
+        client_error = False  # True=客户端请求错误（400/ModelError），应返回 4xx 而非 502
 
         for attempt, k in enumerate(ranked_keys):
             kid = k["key_id"]
@@ -553,6 +554,7 @@ class RouterHandler(BaseHTTPRequestHandler):
                         break
                     if status == 400:
                         log.warning("client error 400, returning to caller without retry")
+                        client_error = True
                         success = False
                         break
                     if status in (401, 403):
@@ -561,6 +563,7 @@ class RouterHandler(BaseHTTPRequestHandler):
                         if "ModelError" in last_error or "model is not" in last_error.lower() \
                            or "not supported" in last_error.lower() or "model not found" in last_error.lower():
                             log.warning("client ModelError (bad model name '%s'), returning 400 without cooling keys", model_name)
+                            client_error = True
                             success = False
                             break
                         mark_key_failed(kid, last_error, cooldown=FAIL_COOLDOWN_SEC)
@@ -590,8 +593,13 @@ class RouterHandler(BaseHTTPRequestHandler):
                     last_error = (f"all keys in cooldown until "
                                   f"{datetime.fromtimestamp(earliest).strftime('%H:%M:%S')}"
                                   f"{' (' + reason_str + ')' if reason_str else ''}")
-            log.error("✗ %s ALL KEYS EXHAUSTED (%.2fs): %s", model_name, elapsed, last_error)
-            self._error_response(502, f"All OCG keys exhausted: {last_error}")
+            if client_error:
+                # 客户端请求错误（400/ModelError）：返回 400，不是服务端故障
+                log.warning("✗ %s client error (%.2fs): %s", model_name, elapsed, last_error)
+                self._error_response(400, last_error)
+            else:
+                log.error("✗ %s ALL KEYS EXHAUSTED (%.2fs): %s", model_name, elapsed, last_error)
+                self._error_response(502, f"All OCG keys exhausted: {last_error}")
 
     def _forward_request(self, api_key, body, is_stream):
         """转发单次请求到指定 key。返回 {ok, status, error}。"""