mqtt切换为远程服务器部署，优化简单指令解析时长

src/cognitive_agent/application/seat_service.py

@@ -171,6 +171,7 @@ MODE_PROFILE_UPDATE_EXPLICIT_TARGET_PATTERN = re.compile(
 )
 WIRED_CAN_TRANSPORT = "wired_can"
 WIRELESS_MQTT_TRANSPORT = "wireless_mqtt"
+LOCAL_CONTROL_BUDGET_MS = 1000
 VEHICLE_CONTROL_TRANSPORTS = frozenset(
     {WIRED_CAN_TRANSPORT, WIRELESS_MQTT_TRANSPORT}
 )
@@ -292,6 +293,7 @@ CONTROL_CLAUSE_SUBJECT_PATTERN = re.compile(
     r"座椅|座位|靠背|椅背|坐盆|坐垫|腿托|脚托|滑轨|"
     r"通风|加热|按摩|零重力|半躺|女王副驾|休闲躺椅|"
     r"遮阳帘|天幕帘|天幕|车窗|窗户|"
+    r"(?:左前|前左|右前|前右|左后|后左|右后|后右)(?:排)?窗|"
     r"灯光|室内灯|车内灯|阅读灯|氛围灯|"
     r"空调|压缩机|温度|风量|风速|内循环|外循环|"
     r"车门|门锁|后备箱|前大灯|近光灯|远光灯|雾灯|转向灯|双闪|"
@@ -302,10 +304,17 @@ CONTROL_CLAUSE_PREDICATE_PATTERN = re.compile(
     r"调整|调节|调到|调成|调为|调至|设置|设为|设到|切换|"
     r"移动|移到|移至|前移|后移|上移|下移|升高|降低|抬起|放倒|"
     r"调直|立直|竖直|归零|复位|锁上|上锁|落锁|解锁|"
-    r"零重力|半躺|最大|最小|最高|最低|最前|最后|到底|"
+    r"零重力|半躺|一半|半开|半关|最大|最小|最高|最低|最前|最后|到底|"
     r"(?:\d+(?:\.\d+)?|[零一二三四五六七八九十百两半]+)"
     r"(?:%|％|度|℃|档|级|秒|毫秒)"
 )
+# These verbs identify an independent request that may be outside the local
+# vehicle registry.  They are used only to split mixed sentences so an
+# already deterministic clause never has to be repeated to a hosted model.
+CLOUD_REQUEST_CLAUSE_PATTERN = re.compile(
+    r"播放|暂停播放|停止播放|导航|搜索|查找|查询|拨打|呼叫|发送|"
+    r"提醒|推荐|讲解|介绍|告诉|规划"
+)
 RESTORE_MODE_PROFILE_PREFIX = re.compile(
     r"^(?:请(?:帮我)?|帮我)?(?:恢复|还原|重置|初始化)"
     r"(?P<name>[^，。,.!！\s]{1,64}?模式)"
@@ -1613,9 +1622,11 @@ class SeatDemoService:
         on_commands_validated: Callable[[dict[str, Any]], None] | None = None,
         assistant_plan: dict[str, Any] | None = None,
         mode_profile_draft_context: dict[str, Any] | None = None,
+        _cloud_instruction_context: dict[str, Any] | None = None,
     ) -> dict[str, Any]:
         """Validate and execute deterministic controls before hosted parsing."""
 
+        local_stage_started = time.perf_counter()
         utterance = utterance.strip()
         if not utterance:
             raise ValueError("utterance must not be empty")
@@ -1644,6 +1655,7 @@ class SeatDemoService:
                 _on_commands_validated=on_commands_validated,
                 _assistant_plan=assistant_plan,
                 _mode_profile_draft_context=mode_profile_draft_context,
+                _cloud_instruction_context=_cloud_instruction_context,
             )
             result.update(
                 {
@@ -1662,14 +1674,71 @@ class SeatDemoService:
             vehicle_profile,
         )
         clause_results = self._local_clause_results(utterance, context)
-        unresolved_clauses = [
+        locally_parsed_clauses = [
+            clause
+            for clause, _, status in clause_results
+            if status != "unparsed"
+        ]
+        locally_resolved_clauses = [
+            clause
+            for clause, _, status in clause_results
+            if status == "complete"
+        ]
+        locally_unresolved_clauses = [
             clause
             for clause, _, status in clause_results
             if status != "complete"
         ]
-        started = time.perf_counter()
+        supplied_routing = (
+            _cloud_instruction_context
+            if isinstance(_cloud_instruction_context, dict)
+            else {}
+        )
+
+        def routing_clauses(key: str) -> list[str]:
+            raw = supplied_routing.get(key)
+            if not isinstance(raw, list):
+                return []
+            return [
+                text
+                for item in raw[:64]
+                for text in [str(item).strip()[:1000]]
+                if text
+            ]
+
+        def unique_clauses(*groups: list[str]) -> list[str]:
+            return list(dict.fromkeys(item for group in groups for item in group))
+
+        parsed_clauses = unique_clauses(
+            routing_clauses("parsed_clauses"),
+            locally_parsed_clauses,
+        )
+        resolved_clauses = unique_clauses(
+            routing_clauses("resolved_clauses"),
+            locally_resolved_clauses,
+        )
+        unresolved_clauses = locally_unresolved_clauses
+        routing_unresolved_clauses = unique_clauses(
+            routing_clauses("unresolved_clauses"),
+            locally_unresolved_clauses,
+        )
+        original_utterance = str(
+            supplied_routing.get("original_utterance") or utterance
+        ).strip()[:1000]
+        natural_language_routing = {
+            "original_utterance": original_utterance,
+            "parsed_clauses": parsed_clauses,
+            "resolved_clauses": resolved_clauses,
+            "unresolved_clauses": routing_unresolved_clauses,
+        }
+        local_dispatch_elapsed_ms: float | None = None
 
         def publish_local_commands(validated: dict[str, Any]) -> None:
+            nonlocal local_dispatch_elapsed_ms
+            local_dispatch_elapsed_ms = round(
+                (time.perf_counter() - local_stage_started) * 1000,
+                3,
+            )
             if on_commands_validated is None:
                 return
             event = deepcopy(validated)
@@ -1681,6 +1750,14 @@ class SeatDemoService:
                     "supplement_required": bool(
                         provider != "local" and unresolved_clauses
                     ),
+                    "natural_language_routing": deepcopy(
+                        natural_language_routing
+                    ),
+                    "local_control_budget_ms": LOCAL_CONTROL_BUDGET_MS,
+                    "local_control_elapsed_ms": local_dispatch_elapsed_ms,
+                    "local_control_within_budget": (
+                        local_dispatch_elapsed_ms < LOCAL_CONTROL_BUDGET_MS
+                    ),
                 }
             )
             on_commands_validated(event)
@@ -1702,8 +1779,17 @@ class SeatDemoService:
             _on_commands_validated=publish_local_commands,
             _assistant_plan=assistant_plan,
             _mode_profile_draft_context=mode_profile_draft_context,
+            _cloud_instruction_context=natural_language_routing,
+        )
+        elapsed_ms = round(
+            (time.perf_counter() - local_stage_started) * 1000,
+            3,
+        )
+        local_control_elapsed_ms = (
+            local_dispatch_elapsed_ms
+            if local_dispatch_elapsed_ms is not None
+            else elapsed_ms
         )
-        elapsed_ms = round((time.perf_counter() - started) * 1000, 3)
         supplement_required = bool(
             provider != "local" and unresolved_clauses
         )
@@ -1713,7 +1799,17 @@ class SeatDemoService:
                 "provider_requested": provider,
                 "preview_request_id": result["request_id"],
                 "preview_elapsed_ms": elapsed_ms,
+                "local_control_budget_ms": LOCAL_CONTROL_BUDGET_MS,
+                "local_control_elapsed_ms": local_control_elapsed_ms,
+                "local_control_within_budget": (
+                    local_control_elapsed_ms < LOCAL_CONTROL_BUDGET_MS
+                ),
+                "parsed_clauses": parsed_clauses,
+                "resolved_clauses": resolved_clauses,
                 "unresolved_clauses": unresolved_clauses,
+                "natural_language_routing": deepcopy(
+                    natural_language_routing
+                ),
                 "supplement_required": supplement_required,
             }
         )
@@ -1737,7 +1833,10 @@ class SeatDemoService:
             "hardware_client_id": hardware_client_id,
             "reasoning_effort": reasoning_effort,
             "control_transport": control_transport,
+            "parsed_clauses": parsed_clauses,
+            "resolved_clauses": resolved_clauses,
             "unresolved_clauses": unresolved_clauses,
+            "natural_language_routing": natural_language_routing,
             "preview_control_signatures": result.get(
                 "parsed_control_signatures", []
             ),
@@ -1802,6 +1901,34 @@ class SeatDemoService:
             for item in manifest.get("preview_control_signatures", [])
             if isinstance(item, list)
         }
+        manifest_routing = manifest.get("natural_language_routing")
+        natural_language_routing = (
+            deepcopy(manifest_routing)
+            if isinstance(manifest_routing, dict)
+            else {
+                "original_utterance": str(manifest.get("utterance") or ""),
+                "parsed_clauses": list(manifest.get("parsed_clauses") or []),
+                "resolved_clauses": list(manifest.get("resolved_clauses") or []),
+                "unresolved_clauses": unresolved_clauses,
+            }
+        )
+        routing_unresolved = natural_language_routing.get("unresolved_clauses")
+        natural_language_routing["unresolved_clauses"] = list(
+            dict.fromkeys(
+                [
+                    str(item).strip()
+                    for item in (
+                        routing_unresolved
+                        if isinstance(routing_unresolved, list)
+                        else []
+                    )
+                    if str(item).strip()
+                ]
+                + unresolved_clauses
+            )
+        )
+        cloud_utterance = "，".join(unresolved_clauses)
+        natural_language_routing["cloud_input_utterance"] = cloud_utterance
 
         def publish_cloud_commands(validated: dict[str, Any]) -> None:
             if on_commands_validated is None:
@@ -1818,7 +1945,7 @@ class SeatDemoService:
             on_commands_validated(event)
 
         supplement = self.process(
-            "，".join(unresolved_clauses),
+            cloud_utterance,
             provider=selected_provider,
             conversation_id=str(manifest["conversation_id"]),
             execution_mode=str(manifest.get("execution_mode") or "dry_run"),
@@ -1837,6 +1964,7 @@ class SeatDemoService:
             _model_parse_source="cloud_supplement",
             _allow_empty_control_result=True,
             _on_commands_validated=publish_cloud_commands,
+            _cloud_instruction_context=natural_language_routing,
         )
         merged = self._merge_native_can_phases(preview, supplement)
         merged.update(
@@ -1853,6 +1981,7 @@ class SeatDemoService:
                 "supplement_request_id": supplement["request_id"],
                 "supplement_required": False,
                 "unresolved_clauses": unresolved_clauses,
+                "natural_language_routing": natural_language_routing,
             }
         )
         self._persist(merged)
@@ -2149,6 +2278,7 @@ class SeatDemoService:
         _speech_understanding: dict[str, Any] | None = None,
         _assistant_plan: dict[str, Any] | None = None,
         _mode_profile_draft_context: dict[str, Any] | None = None,
+        _cloud_instruction_context: dict[str, Any] | None = None,
     ) -> dict[str, Any]:
         control_transport = self._validate_control_transport(control_transport)
         # Creating a reusable scene is a configuration workflow.  Synthesize
@@ -2197,6 +2327,7 @@ class SeatDemoService:
                 _speech_understanding,
                 assistant_plan=_assistant_plan,
                 mode_profile_draft_context=_mode_profile_draft_context,
+                cloud_instruction_context=_cloud_instruction_context,
             )
         if control_transport == WIRELESS_MQTT_TRANSPORT:
             # The head unit owns both physical CAN channels in wireless mode.
@@ -2222,6 +2353,7 @@ class SeatDemoService:
                 _speech_understanding,
                 assistant_plan=_assistant_plan,
                 mode_profile_draft_context=_mode_profile_draft_context,
+                cloud_instruction_context=_cloud_instruction_context,
             )
         if hardware_access_bypass or self._driver_only_vehicle_execution_active(
             control_transport
@@ -2247,6 +2379,7 @@ class SeatDemoService:
                     _speech_understanding,
                     assistant_plan=_assistant_plan,
                     mode_profile_draft_context=_mode_profile_draft_context,
+                    cloud_instruction_context=_cloud_instruction_context,
                 )
         authorized_client_id = self.hardware_access.ensure_authorized(hardware_client_id)
         with self.hardware_execution.reserve():
@@ -2271,6 +2404,7 @@ class SeatDemoService:
                     _speech_understanding,
                     assistant_plan=_assistant_plan,
                     mode_profile_draft_context=_mode_profile_draft_context,
+                    cloud_instruction_context=_cloud_instruction_context,
                 )
 
     def _process_once(
@@ -2296,6 +2430,7 @@ class SeatDemoService:
         allow_fuzzy_mode_profile_match: bool = True,
         allow_scene_composition: bool = False,
         mode_profile_draft_context: dict[str, Any] | None = None,
+        cloud_instruction_context: dict[str, Any] | None = None,
     ) -> dict[str, Any]:
         control_transport = self._validate_control_transport(control_transport)
         utterance = utterance.strip()
@@ -2626,6 +2761,7 @@ class SeatDemoService:
                 model_parse_source=model_parse_source,
                 assistant_plan=assistant_plan,
                 allow_scene_composition=allow_scene_composition,
+                cloud_instruction_context=cloud_instruction_context,
             )
             dynamic_mode_stop = self._is_dynamic_mode_stop_request(utterance)
             if dynamic_mode_stop:
@@ -7280,6 +7416,11 @@ class SeatDemoService:
                 and CONTROL_CLAUSE_PREDICATE_PATTERN.search(value)
             )
 
+        def looks_like_independent_request(value: str) -> bool:
+            return looks_like_control(value) or bool(
+                CLOUD_REQUEST_CLAUSE_PATTERN.search(value)
+            )
+
         def split_independent_controls(
             value: str,
             separator_pattern: str,
@@ -7290,7 +7431,14 @@ class SeatDemoService:
             for index in range(1, len(tokens), 2):
                 separator = tokens[index]
                 next_item = tokens[index + 1].strip()
-                if looks_like_control(current) and looks_like_control(next_item):
+                if (
+                    looks_like_independent_request(current)
+                    and looks_like_independent_request(next_item)
+                    and (
+                        looks_like_control(current)
+                        or looks_like_control(next_item)
+                    )
+                ):
                     if current:
                         separated.append(current)
                     current = next_item
@@ -7600,6 +7748,7 @@ class SeatDemoService:
         model_parse_source: str = "cloud_model",
         assistant_plan: dict[str, Any] | None = None,
         allow_scene_composition: bool = False,
+        cloud_instruction_context: dict[str, Any] | None = None,
     ) -> tuple[dict[str, Any], str, str | None]:
         vehicle_profile = vehicle_profile or self.vehicle_profiles.default()
         if assistant_plan is not None and allow_scene_composition:
@@ -7764,9 +7913,29 @@ class SeatDemoService:
             interpret_options = {}
             if reasoning_effort is not None:
                 interpret_options["reasoning_effort"] = reasoning_effort
+            hosted_context = deepcopy(context)
+            if isinstance(cloud_instruction_context, dict):
+                routing_context = deepcopy(cloud_instruction_context)
+            else:
+                routing_context = {
+                    "original_utterance": utterance,
+                    "parsed_clauses": [
+                        clause
+                        for clause, _, status in clause_results
+                        if status != "unparsed"
+                    ],
+                    "resolved_clauses": [
+                        clause
+                        for clause, _, status in clause_results
+                        if status == "complete"
+                    ],
+                    "unresolved_clauses": fallback_clauses,
+                }
+            routing_context["cloud_input_utterance"] = fallback_utterance
+            hosted_context["natural_language_routing"] = routing_context
             plan = self.openai.interpret(
                 fallback_utterance,
-                context,
+                hosted_context,
                 self.capability_context,
                 **interpret_options,
             )