"""Workflow-editor helpers for config routes.""" import copy import json import time from typing import Optional, Dict, Any, Callable, get_args from fastapi import HTTPException from app.core.config import get_logger, WorkflowCancelledError from app.core.page_lifecycle import BACKGROUND_WAKE_CDP_TIMEOUT, BACKGROUND_WAKE_JS_TIMEOUT from app.models.schemas import ActionType from app.core.workflow.flow_runtime import (FlowProgram, FlowValidationError, validate_workflow, has_control_flow, capture_page_state, validate_inputs) from app.services.config_engine import config_engine from app.utils.site_url import extract_remote_site_domain from app.core.driver import driver_for_tab logger = get_logger('API.CONFIG.WORKFLOW') # 修复:保存时校验动作名。合法集合直接取自 schemas.ActionType; # 运行期 executor 的 dispatch 用原始 action 精确匹配大写字面量、不做归一化,故此处也要求精确大写。 VALID_WORKFLOW_ACTIONS = frozenset(get_args(ActionType)) def _notify_workflow_editor_action_result(tab, action_id: str, success: bool, message: str, trace=None) -> None: """将测试结果回推给已注入的可视化编辑器页面。""" try: driver_for_tab(tab).run_js( """ return (function(actionId, ok, text, trace) { if (window.WorkflowEditor && typeof window.WorkflowEditor.handleBackendResult === 'function') { window.WorkflowEditor.handleBackendResult(actionId, ok, text, trace); return true; } return false; })(arguments[0], arguments[1], arguments[2], arguments[3]); """, str(action_id or ""), bool(success), str(message or ""), list(trace or [])[:500], ) except Exception as e: logger.debug(f"回推编辑器测试结果失败(忽略): {e}") def _notify_workflow_editor_action_status(tab, action_id: str, phase: str, message: str) -> None: """将测试中间状态回推给已注入的可视化编辑器页面。""" try: driver_for_tab(tab).run_js( """ return (function(actionId, phaseName, text) { if (window.WorkflowEditor && typeof window.WorkflowEditor.handleBackendStatus === 'function') { window.WorkflowEditor.handleBackendStatus(actionId, phaseName, text); return true; } return false; })(arguments[0], arguments[1], arguments[2]); """, str(action_id or ""), str(phase or ""), str(message or ""), ) except Exception as e: logger.debug(f"回推编辑器测试状态失败(忽略): {e}") def _wake_workflow_editor_test_tab(session) -> None: """在测试前尽量唤醒后台标签页,避免页面被冻结导致点击无效。""" if session is None: return try: if hasattr(session, "activate"): session.activate() except Exception: pass focus_emulation_set = False try: driver_for_tab(session.tab).run_cdp( "Emulation.setFocusEmulationEnabled", enabled=True, _timeout=BACKGROUND_WAKE_CDP_TIMEOUT, ) focus_emulation_set = True except Exception: pass try: driver_for_tab(session.tab).run_cdp( "Page.setWebLifecycleState", state="active", _timeout=BACKGROUND_WAKE_CDP_TIMEOUT, ) except Exception: pass try: driver_for_tab(session.tab).run_js("return document.readyState || '';", timeout=BACKGROUND_WAKE_JS_TIMEOUT) except Exception: pass finally: if focus_emulation_set: try: driver_for_tab(session.tab).run_cdp( "Emulation.setFocusEmulationEnabled", enabled=False, _timeout=BACKGROUND_WAKE_CDP_TIMEOUT, ) except Exception: pass def _execute_workflow_editor_test_payload( browser_instance, data: Dict[str, Any], progress_callback: Optional[Callable[[str, str], None]] = None ) -> Dict[str, Any]: """复用真实执行器执行可视化编辑器测试。""" from app.core.workflow.executor import WorkflowExecutor action_labels = { "CLICK": "点击元素", "SELECT_MODEL": "选择请求模型", "COORD_CLICK": "坐标点击", "COORD_SCROLL": "模拟滑动", "FILL_INPUT": "填入内容", "STREAM_WAIT": "流式等待", # 修复:补上缺失的两项,否则测试进度提示会退化成显示英文原始 action 名 "STREAM_OUTPUT": "流式等待", "WAIT": "等待", "KEY_PRESS": "按键", "JS_EXEC": "执行脚本", "PAGE_FETCH": "页面直发", "READONLY_HINT": "只读提示", } domain = str(data.get("domain") or "").strip() tab_id = str(data.get("tab_id") or "").strip() workflow = data.get("workflow") or [] selectors = data.get("selectors") or {} preset_name = str(data.get("preset_name") or "").strip() or None prompt_text = str(data.get("prompt") or "") stealth = bool(data.get("stealth", False)) stream_config = dict(data.get("stream_config") or {}) image_config = data.get("image_extraction") or {} file_paste_config = data.get("file_paste") or {} if not domain: raise HTTPException(status_code=400, detail="缺少 domain") if not isinstance(workflow, list) or not workflow: raise HTTPException(status_code=400, detail="workflow 必须是非空数组") if not isinstance(selectors, dict): raise HTTPException(status_code=400, detail="selectors 必须是对象") try: validate_workflow(workflow) validate_inputs(data.get("workflow_variables", {})) except FlowValidationError as exc: raise HTTPException(status_code=400, detail=str(exc)) from exc if not browser_instance.get_browser_handle(): raise HTTPException(status_code=503, detail="浏览器未连接") task_id = f"workflow_editor_test_{time.time_ns()}" session = None executor = None flow_program = None started_at = time.time() executed = 0 try: if tab_id and getattr(browser_instance, "tab_pool", None) is not None: session = browser_instance.tab_pool.acquire_by_raw_tab_id( tab_id, task_id, timeout=5, count_request=False, ) tab = session.tab if session is not None else ( browser_instance.get_tab(tab_id) if tab_id else browser_instance.get_latest_tab() ) logger.debug( "[WFE_TEST] tab resolved " f"session={getattr(session, 'id', None)!r} " f"raw_tab_id={getattr(tab, 'tab_id', None)!r} " f"url={str(getattr(tab, 'url', '') or '')[:160]!r}" ) test_stream_config = dict(stream_config or {}) test_network_config = dict(test_stream_config.get("network") or {}) test_stream_config["hard_timeout"] = min( float(test_stream_config.get("hard_timeout", 45) or 45), 45.0, ) test_network_config["first_response_timeout"] = min( float(test_network_config.get("first_response_timeout", 12) or 12), 12.0, ) test_network_config["silence_threshold"] = max( 0.5, min(float(test_network_config.get("silence_threshold", 3) or 3), 30.0), ) test_stream_config["network"] = test_network_config logger.debug( "[WFE_TEST] start " f"domain={domain!r} tab_id={tab_id!r} preset={preset_name!r} " f"steps={len(workflow)} selectors={len(selectors)} " f"hard_timeout={test_stream_config['hard_timeout']:.1f}s " f"first_response_timeout={test_network_config['first_response_timeout']:.1f}s " f"silence_threshold={test_network_config['silence_threshold']:.1f}s" ) if progress_callback is None and data.get("visual_feedback") is True: progress_callback = lambda phase, message: _notify_workflow_editor_action_status(tab, "", phase, message) if progress_callback: progress_callback("running", f"本地控制台已接管,准备执行 {len(workflow)} 个动作") _wake_workflow_editor_test_tab(session) url = str(getattr(tab, "url", "") or "") actual_domain = extract_remote_site_domain(url) if not actual_domain: raise HTTPException(status_code=400, detail="当前页面不是可解析的网站") if actual_domain != domain: raise HTTPException( status_code=400, detail=f"域名不匹配:当前页面是 {actual_domain},测试目标是 {domain}" ) resolved_site_config = config_engine.get_site_config( domain, html_content=lambda: getattr(tab, "html", ""), preset_name=preset_name, ) or {} site_advanced_config = config_engine.get_site_advanced_config( domain, preset_name=preset_name, ) extractor = config_engine.get_site_extractor(domain, preset_name=preset_name) cancellation = {"checked_at": 0.0, "cancelled": False} def editor_should_stop(): if time.time() - started_at > 45: return True if data.get("visual_feedback") is True and time.monotonic() - cancellation["checked_at"] >= 0.2: cancellation["checked_at"] = time.monotonic() try: cancellation["cancelled"] = cancellation["cancelled"] or driver_for_tab(tab).run_js( "return window.__WORKFLOW_EDITOR_TEST_CANCELLED__ === true || (arguments[0] && window.__WORKFLOW_EDITOR_TEST_RUN_ID__ !== arguments[0]);", str(data.get("visual_run_id") or "")[:128], timeout=0.5, ) is True except Exception: pass # Browser failure still respects the overall time budget. return cancellation["cancelled"] executor = WorkflowExecutor( tab=tab, stealth_mode=stealth, should_stop_checker=editor_should_stop, extractor=extractor, image_config=image_config, stream_config=test_stream_config or resolved_site_config.get("stream_config") or {}, file_paste_config=file_paste_config, site_advanced_config=site_advanced_config, selectors=selectors, session=session, ) executed = 0 context = { "prompt": prompt_text, "images": [], } # 修复:此前不透传 model/model_catalog,导致 SELECT_MODEL 步骤在编辑器测试里必然 early return, # 但仍计入 executed_steps,测试显示“成功”而实际未执行。 requested_model = str((data or {}).get("model") or "").strip() if requested_model: context["model"] = requested_model model_catalog = (data or {}).get("model_catalog") if isinstance(model_catalog, dict) and model_catalog: context["model_catalog"] = model_catalog skipped_select_model = 0 flow_program = None if has_control_flow(workflow) or data.get("workflow_variables"): flow_program = FlowProgram( workflow, context=context, inputs=data.get("workflow_variables"), capture=lambda spec, target: capture_page_state(executor, spec, target, flow_program.variables), stop_checker=executor._check_cancelled, deadline_seconds=45, ) workflow = flow_program.steps with executor.workflow_execution_scope(): step_index = 0 while step_index < len(workflow): if flow_program is not None: step_index = flow_program.next_leaf(step_index) if step_index >= len(workflow): break try: step, flow_selector = flow_program.prepare_leaf(step_index, selectors) except Exception as exc: resumed = flow_program.recover(exc, min(step_index, len(workflow) - 1)) if resumed is None: raise step_index = resumed continue else: step = workflow[step_index] action = str(step.get("action") or "").strip() target_key = str(step.get("target") or "") optional = bool(step.get("optional", False)) value = step.get("value") execution_policy = step.get("execution") selector = flow_selector if flow_program is not None else selectors.get(target_key, "") current_index = step_index + 1 logger.debug( "[WFE_TEST] step " f"index={current_index} action={action!r} target={target_key!r} " f"selector={selector!r} optional={optional}" ) if not action: raise HTTPException(status_code=400, detail="workflow 中存在缺少 action 的步骤") # 修复:未指定模型时 SELECT_MODEL 必然 early return(executor_actions.py:1284-1289), # 此前仍会计入 executed_steps,让"已测试 N 个步骤"虚高、掩盖空跑。 # 注意:这是 while 循环,跳过前必须自行推进 step_index,否则死循环。 if action == "SELECT_MODEL" and not requested_model: skipped_select_model += 1 if progress_callback: progress_callback( "step", f"跳过 {current_index}/{len(workflow)} · 选择请求模型(未指定测试模型名)" ) step_index += 1 continue if progress_callback: # Ephemeral page feedback uses resolved targets; it is NOT # persisted in the privacy-preserving workflow trace. phase = "step" if data.get("visual_feedback") is True: phase = "step:" + json.dumps({ "path": step.get("_path", f"root.{step_index}"), "selector": selector, "coordinates": value if action in {"COORD_CLICK", "COORD_SCROLL"} and isinstance(value, dict) else None, }, ensure_ascii=False) progress_callback( phase, f"执行 {current_index}/{len(workflow)} · {action_labels.get(action, action)}" ) if data.get("visual_feedback") is True: # Give the browser one visible frame before performing # the action. Only editor tests opt into this pacing. until = time.monotonic() + 0.25 while time.monotonic() < until and not executor._check_cancelled(): time.sleep(0.02) if flow_program is None and action == "FILL_INPUT" and value is not None: context["prompt"] = str(value) step_events = executor.execute_step( action=action, selector=selector, target_key=target_key, value=value, optional=optional, context=flow_program.leaf_context(step) if flow_program is not None else context, execution=execution_policy, ) if flow_program is not None: step_events = flow_program.events(step_events, step_index) try: for _ in step_events: pass except Exception as exc: if flow_program is None: raise resumed = flow_program.recover(exc, step_index) if resumed is None: raise step_index = resumed continue if executor._check_cancelled(): raise WorkflowCancelledError("workflow_cancelled") executed += 1 if ( action == "PAGE_FETCH" and hasattr(executor, "consume_last_request_transport_sent") and executor.consume_last_request_transport_sent() ): step_index = executor._consume_request_transport_followup_steps( workflow, step_index, ) step_index += 1 logger.debug( "[WFE_TEST] done " f"domain={domain!r} tab_id={tab_id or str(getattr(tab, 'tab_id', '') or '')!r} " f"executed_steps={executed} duration={time.time() - started_at:.2f}s" ) # 修复:把被跳过的 SELECT_MODEL 步骤如实告知前端,避免“已测试 N 个步骤”掩盖空跑 message = f"已测试 {executed} 个步骤" if skipped_select_model: message += f"(其中 {skipped_select_model} 个选择模型步骤已跳过:未指定测试模型名)" return { "success": True, "message": message, "domain": domain, "tab_id": tab_id or str(getattr(tab, "tab_id", "") or ""), "preset_name": preset_name or config_engine.get_default_preset(domain) or "主预设", "executed_steps": executed, "trace": flow_program.trace if flow_program is not None else [], "skipped_select_model_steps": skipped_select_model, "_tab_ref": tab, } except Exception as exc: if data.get("visual_feedback") is True and isinstance(exc, WorkflowCancelledError): return {"success": False, "cancelled": True, "message": "测试已停止(取消或达到时间预算),已执行的网页操作不会撤销。", "trace": flow_program.trace if flow_program is not None else [], "executed_steps": executed, "_tab_ref": tab} if flow_program is None: raise return {"success": False, "message": str(exc), "trace": flow_program.trace, "executed_steps": executed, "_tab_ref": tab} finally: if executor is not None: try: executor.cleanup_after_workflow() except Exception as e: logger.debug(f"[WFE_TEST] executor cleanup skipped: {e}") if session is not None and getattr(browser_instance, "tab_pool", None) is not None: logger.debug(f"[WFE_TEST] release session={session.id!r}") browser_instance.tab_pool.release( session.id, check_triggers=False, expected_task_id=task_id, ) def _save_site_workflow_payload(domain: str, data: Dict[str, Any]) -> Dict[str, Any]: """保存工作流配置,供直接 API 和桥接模式复用。""" new_workflow = data.get("workflow") new_selectors = data.get("selectors") preset_name = data.get("preset_name") if new_workflow is None: raise HTTPException(status_code=400, detail="缺少 workflow 字段") if not isinstance(new_workflow, list): raise HTTPException(status_code=400, detail="workflow 必须是数组") # 修复:此前不校验动作名,拼错(如 STREAM_WIAT)或小写(click)会静默入库, # 运行期只在 debug 日志里记一行“未知动作”,表现为工作流跑完但什么都没发生。 for i, step in enumerate(new_workflow): if not isinstance(step, dict): raise HTTPException(status_code=400, detail=f"第 {i + 1} 步必须是对象") action = step.get("action") if not isinstance(action, str) or action not in VALID_WORKFLOW_ACTIONS: raise HTTPException( status_code=400, detail=( f"第 {i + 1} 步的动作类型无效: {action}" f"(动作名必须大写,可选值:{'、'.join(sorted(VALID_WORKFLOW_ACTIONS))})" ), ) try: validate_workflow(new_workflow) except FlowValidationError as exc: raise HTTPException(status_code=400, detail=str(exc)) from exc if new_selectors is not None and ( not isinstance(new_selectors, dict) or isinstance(new_selectors, list) ): raise HTTPException(status_code=400, detail="selectors 必须是对象") config_engine.refresh_if_changed() if domain not in config_engine.sites: raise HTTPException(status_code=404, detail=f"站点不存在: {domain}") site = config_engine.sites[domain] presets = site.get("presets", {}) resolved_preset_name = None if preset_name: resolved_preset_name = config_engine._resolve_preset_alias_key(preset_name, presets) if resolved_preset_name not in presets: raise HTTPException(status_code=404, detail=f"预设不存在: {preset_name}") preset_data = config_engine._get_site_data(domain, resolved_preset_name) if preset_data is None: raise HTTPException(status_code=404, detail="站点或预设不存在") workflow_existed = "workflow" in preset_data previous_workflow = copy.deepcopy(preset_data.get("workflow")) selectors_existed = "selectors" in preset_data previous_selectors = copy.deepcopy(preset_data.get("selectors")) preset_data["workflow"] = new_workflow if new_selectors is not None: preset_data["selectors"] = new_selectors def rollback_workflow_memory() -> None: if workflow_existed: preset_data["workflow"] = previous_workflow else: preset_data.pop("workflow", None) if new_selectors is not None: if selectors_existed: preset_data["selectors"] = previous_selectors else: preset_data.pop("selectors", None) try: success = config_engine.save_config() except Exception as exc: rollback_workflow_memory() raise HTTPException(status_code=500, detail=f"保存配置文件失败: {exc}") from exc if not success: rollback_workflow_memory() raise HTTPException(status_code=500, detail="保存配置文件失败") used_preset = ( resolved_preset_name or config_engine.get_default_preset(domain) or "主预设" ) logger.info(f"站点 {domain} [{used_preset}] 工作流已更新: {len(new_workflow)} 个步骤") return { "status": "success", "message": f"工作流已保存", "domain": domain, "preset_name": used_preset, "steps_count": len(new_workflow) } def _scan_workflow_scripts() -> Dict[str, Any]: """扫描可用工作流脚本列表""" from app.core.workflow.script_loader import script_loader scripts = script_loader.scan_available_scripts() return {"scripts": scripts} def _load_workflow_script_content(target_path: str) -> Dict[str, Any]: """加载指定脚本的内容与元数据(用于前端预览)""" from app.core.workflow.script_loader import script_loader from app.core.config import WorkflowError if not target_path: raise HTTPException(status_code=400, detail="缺少脚本路径参数") try: resolved_path = script_loader.resolve_script_path(target_path) content = script_loader.load_script_content(resolved_path) stat = resolved_path.stat() mtime = int(stat.st_mtime) desc = script_loader.extract_script_description(content, resolved_path.name) rel_path = resolved_path.relative_to(script_loader.base_dir).as_posix() return { "path": rel_path, "uri": f"file://{rel_path}", "name": resolved_path.name, "description": desc, "content": content, "mtime": mtime, } except WorkflowError as e: raise HTTPException(status_code=400, detail=str(e)) except Exception as e: logger.error(f"读取脚本内容失败 ({target_path!r}): {e}") raise HTTPException(status_code=500, detail=f"读取脚本失败: {e}") # ================= 请求模型 ================= __all__ = [ '_notify_workflow_editor_action_result', '_notify_workflow_editor_action_status', '_execute_workflow_editor_test_payload', '_save_site_workflow_payload', '_scan_workflow_scripts', '_load_workflow_script_content', ]