From 877bcc2fad3c2faafa2e7ba9b66a110a4a563c99 Mon Sep 17 00:00:00 2001 From: JOJO <1498581755@qq.com> Date: Fri, 6 Mar 2026 17:02:19 +0800 Subject: [PATCH] fix: improve api error diagnostics and raise model quotas --- config/limits.py | 8 +-- modules/usage_tracker.py | 4 +- server/_conversation_segment.py | 22 ++++-- server/chat_flow.py | 88 +++++++++++++++++++---- static/src/composables/useLegacySocket.ts | 22 +++++- utils/api_client.py | 25 ++++--- 6 files changed, 133 insertions(+), 36 deletions(-) diff --git a/config/limits.py b/config/limits.py index 0b2481b..943717b 100644 --- a/config/limits.py +++ b/config/limits.py @@ -13,12 +13,12 @@ CODE_EXECUTION_TIMEOUT = 60 TERMINAL_COMMAND_TIMEOUT = 30 SEARCH_MAX_RESULTS = 10 -# 自动修复与工具调用限制 +# 自动修复与工具调用限制（None 表示不限制） AUTO_FIX_TOOL_CALL = False AUTO_FIX_MAX_ATTEMPTS = 3 -MAX_ITERATIONS_PER_TASK = 100 -MAX_CONSECUTIVE_SAME_TOOL = 50 -MAX_TOTAL_TOOL_CALLS = 100 +MAX_ITERATIONS_PER_TASK = None +MAX_CONSECUTIVE_SAME_TOOL = None +MAX_TOTAL_TOOL_CALLS = None TOOL_CALL_COOLDOWN = 0.5 THINKING_FAST_INTERVAL = 10 diff --git a/modules/usage_tracker.py b/modules/usage_tracker.py index 5377b2f..3276d50 100644 --- a/modules/usage_tracker.py +++ b/modules/usage_tracker.py @@ -12,8 +12,8 @@ QuotaKey = Literal["fast", "thinking", "search"] QUOTA_DEFAULTS = { "default": { - "fast": {"limit": 50, "window_hours": 5}, - "thinking": {"limit": 20, "window_hours": 5}, + "fast": {"limit": 200, "window_hours": 5}, + "thinking": {"limit": 200, "window_hours": 5}, "search": {"limit": 20, "window_hours": 24}, }, "search_daily": {"limit": 20, "window_hours": 24}, diff --git a/server/_conversation_segment.py b/server/_conversation_segment.py index a2fbbd9..88f2dbe 100644 --- a/server/_conversation_segment.py +++ b/server/_conversation_segment.py @@ -695,7 +695,7 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac last_tool_call_time = 0 detected_tool_intent: Dict[str, str] = {} - # 设置最大迭代次数 + # 设置最大迭代次数；None 表示不限制 max_iterations = MAX_ITERATIONS_PER_TASK pending_append = None # {"path": str, "tool_call_id": str, "buffer": str, ...} @@ -1337,12 +1337,16 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac }) maybe_mark_failure_from_message(web_terminal, message) - for iteration in range(max_iterations): + iteration = 0 + while max_iterations is None or iteration < max_iterations: + current_iteration = iteration + 1 + iteration += 1 total_iterations += 1 - debug_log(f"\n--- 迭代 {iteration + 1}/{max_iterations} 开始 ---") + iteration_limit_label = max_iterations if max_iterations is not None else "∞" + debug_log(f"\n--- 迭代 {current_iteration}/{iteration_limit_label} 开始 ---") # 检查是否超过总工具调用限制 - if total_tool_calls >= MAX_TOTAL_TOOL_CALLS: + if MAX_TOTAL_TOOL_CALLS is not None and total_tool_calls >= MAX_TOTAL_TOOL_CALLS: debug_log(f"已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})") sender('system_message', { 'content': f'⚠️ 已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})，任务结束。' @@ -1403,7 +1407,8 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac }) return - print(f"[API] 第{iteration + 1}次调用 (总工具调用: {total_tool_calls}/{MAX_TOTAL_TOOL_CALLS})") + tool_call_limit_label = MAX_TOTAL_TOOL_CALLS if MAX_TOTAL_TOOL_CALLS is not None else "∞" + print(f"[API] 第{current_iteration}次调用 (总工具调用: {total_tool_calls}/{tool_call_limit_label})") # 收集流式响应 async for chunk in web_terminal.api_client.chat(messages, tools, stream=True): @@ -1677,7 +1682,7 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac text_chunk_index += 1 log_backend_chunk( conversation_id, - iteration + 1, + current_iteration, text_chunk_index, elapsed, len(content), @@ -2099,7 +2104,10 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac if tool_name == last_tool_name: consecutive_same_tool[tool_name] += 1 - if consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL: + if ( + MAX_CONSECUTIVE_SAME_TOOL is not None + and consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL + ): debug_log(f"警告: 连续调用相同工具 {tool_name} 已达 {MAX_CONSECUTIVE_SAME_TOOL} 次") sender('system_message', { 'content': f'⚠️ 检测到重复调用 {tool_name} 工具 {MAX_CONSECUTIVE_SAME_TOOL} 次，可能存在循环。' diff --git a/server/chat_flow.py b/server/chat_flow.py index 0b26fb3..03c82c2 100644 --- a/server/chat_flow.py +++ b/server/chat_flow.py @@ -48,7 +48,7 @@ from core.web_terminal import WebTerminal from utils.tool_result_formatter import format_tool_result_for_context from utils.conversation_manager import ConversationManager from utils.api_client import DeepSeekClient -from config.model_profiles import get_model_context_window +from config.model_profiles import get_model_context_window, get_model_profile from .auth_helpers import api_login_required, resolve_admin_policy, get_current_user_record, get_current_username from .context import with_terminal, get_gui_manager, get_upload_guard, build_upload_error_response, ensure_conversation_loaded, reset_system_state, get_user_resources, get_or_create_usage_tracker @@ -425,7 +425,12 @@ def process_message_task(terminal: WebTerminal, message: str, images, sender, cl debug_log(f"任务处理错误: {e}") import traceback traceback.print_exc() - sender('error', {'message': str(e)}) + sender('error', { + 'message': str(e), + 'conversation_id': getattr(getattr(terminal, "context_manager", None), "current_conversation_id", None), + 'task_id': getattr(terminal, "task_id", None) or client_sid, + 'client_sid': client_sid + }) sender('task_complete', { 'total_iterations': 0, 'total_tool_calls': 0, @@ -469,6 +474,25 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac web_terminal = terminal conversation_id = getattr(web_terminal.context_manager, "current_conversation_id", None) videos = videos or [] + raw_sender = sender + + def sender(event_type, data): + """为关键事件补充会话标识，便于前端定位报错归属。""" + if not isinstance(data, dict): + raw_sender(event_type, data) + return + payload = data + if event_type in {"error", "quota_exceeded", "task_stopped", "task_complete"}: + payload = dict(data) + current_conv = conversation_id or getattr(web_terminal.context_manager, "current_conversation_id", None) + if current_conv: + payload.setdefault("conversation_id", current_conv) + task_id = getattr(web_terminal, "task_id", None) or client_sid + if task_id: + payload.setdefault("task_id", task_id) + if client_sid: + payload.setdefault("client_sid", client_sid) + raw_sender(event_type, payload) # 如果是思考模式，重置状态 if web_terminal.thinking_mode: @@ -562,8 +586,9 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac last_tool_call_time = 0 detected_tool_intent: Dict[str, str] = {} - # 设置最大迭代次数（API 可覆盖） - max_iterations = getattr(web_terminal, "max_iterations_override", None) or MAX_ITERATIONS_PER_TASK + # 设置最大迭代次数（API 可覆盖）；None 表示不限制 + max_iterations_override = getattr(web_terminal, "max_iterations_override", None) + max_iterations = max_iterations_override if max_iterations_override is not None else MAX_ITERATIONS_PER_TASK max_api_retries = 4 retry_delay_seconds = 10 @@ -1225,12 +1250,16 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac await asyncio.sleep(0.2) return False - for iteration in range(max_iterations): + iteration = 0 + while max_iterations is None or iteration < max_iterations: + current_iteration = iteration + 1 + iteration += 1 total_iterations += 1 - debug_log(f"\n--- 迭代 {iteration + 1}/{max_iterations} 开始 ---") + iteration_limit_label = max_iterations if max_iterations is not None else "∞" + debug_log(f"\n--- 迭代 {current_iteration}/{iteration_limit_label} 开始 ---") # 检查是否超过总工具调用限制 - if total_tool_calls >= MAX_TOTAL_TOOL_CALLS: + if MAX_TOTAL_TOOL_CALLS is not None and total_tool_calls >= MAX_TOTAL_TOOL_CALLS: debug_log(f"已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})") sender('system_message', { 'content': f'⚠️ 已达到最大工具调用次数限制 ({MAX_TOTAL_TOOL_CALLS})，任务结束。' @@ -1317,7 +1346,8 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac }) return - print(f"[API] 第{iteration + 1}次调用 (总工具调用: {total_tool_calls}/{MAX_TOTAL_TOOL_CALLS})") + tool_call_limit_label = MAX_TOTAL_TOOL_CALLS if MAX_TOTAL_TOOL_CALLS is not None else "∞" + print(f"[API] 第{current_iteration}次调用 (总工具调用: {total_tool_calls}/{tool_call_limit_label})") api_error = None for api_attempt in range(max_api_retries + 1): @@ -1612,7 +1642,7 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac text_chunk_index += 1 log_backend_chunk( conversation_id, - iteration + 1, + current_iteration, text_chunk_index, elapsed, len(content), @@ -1746,15 +1776,39 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac if api_error: + try: + debug_log(f"API错误原始数据: {json.dumps(api_error, ensure_ascii=False)}") + except Exception: + debug_log(f"API错误原始数据(不可序列化): {repr(api_error)}") error_message = "" error_status = None error_type = None + error_code = None + error_text = "" + request_dump = None + error_base_url = None + error_model_id = None if isinstance(api_error, dict): error_status = api_error.get("status_code") - error_type = api_error.get("error_type") - error_message = api_error.get("error_message") or api_error.get("error_text") or "" + error_type = api_error.get("error_type") or api_error.get("type") + error_code = api_error.get("error_code") or api_error.get("code") + error_text = api_error.get("error_text") or "" + error_message = ( + api_error.get("error_message") + or api_error.get("message") + or error_text + or "" + ) + request_dump = api_error.get("request_dump") + error_base_url = api_error.get("base_url") + error_model_id = api_error.get("model_id") + elif isinstance(api_error, str): + error_message = api_error if not error_message: - error_message = "API 请求失败" + if error_status: + error_message = f"API 请求失败（HTTP {error_status}）" + else: + error_message = "API 请求失败" # 若命中阿里云配额错误，立即写入状态并切换到官方 API try: from utils.aliyun_fallback import compute_disabled_until, set_disabled_until @@ -1777,6 +1831,11 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac 'message': error_message, 'status_code': error_status, 'error_type': error_type, + 'error_code': error_code, + 'error_text': error_text, + 'request_dump': request_dump, + 'base_url': error_base_url, + 'model_id': error_model_id, 'retry': bool(can_retry), 'retry_in': retry_delay_seconds if can_retry else None, 'attempt': api_attempt + 1, @@ -2092,7 +2151,10 @@ async def handle_task_with_sender(terminal: WebTerminal, workspace: UserWorkspac if tool_name == last_tool_name: consecutive_same_tool[tool_name] += 1 - if consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL: + if ( + MAX_CONSECUTIVE_SAME_TOOL is not None + and consecutive_same_tool[tool_name] >= MAX_CONSECUTIVE_SAME_TOOL + ): debug_log(f"警告: 连续调用相同工具 {tool_name} 已达 {MAX_CONSECUTIVE_SAME_TOOL} 次") sender('system_message', { 'content': f'⚠️ 检测到重复调用 {tool_name} 工具 {MAX_CONSECUTIVE_SAME_TOOL} 次，可能存在循环。' diff --git a/static/src/composables/useLegacySocket.ts b/static/src/composables/useLegacySocket.ts index b646354..28f2719 100644 --- a/static/src/composables/useLegacySocket.ts +++ b/static/src/composables/useLegacySocket.ts @@ -1409,14 +1409,28 @@ export async function initializeLegacySocket(ctx: any) { const msg = data?.message || '发生未知错误'; const code = data?.status_code; const errType = data?.error_type; + const errCode = data?.error_code; + const dumpPath = data?.request_dump; + const baseUrl = data?.base_url; + const modelId = data?.model_id; + const conversationId = data?.conversation_id; + const taskId = data?.task_id; const shouldRetry = Boolean(data?.retry); const retryIn = Number(data?.retry_in) || 5; const retryAttempt = Number(data?.attempt) || 1; const retryMax = Number(data?.max_attempts) || retryAttempt; + const detailParts = [ + dumpPath ? `请求记录: ${dumpPath}` : '', + baseUrl ? `接口: ${baseUrl}` : '', + modelId ? `模型: ${modelId}` : '', + conversationId ? `对话ID: ${conversationId}` : '', + taskId ? `任务ID: ${taskId}` : '' + ].filter(Boolean); + const detailText = detailParts.length ? `\n${detailParts.join('\n')}` : ''; if (typeof ctx.uiPushToast === 'function') { ctx.uiPushToast({ title: code ? `API错误 ${code}` : 'API错误', - message: errType ? `${errType}: ${msg}` : msg, + message: `${errType ? `${errType}${errCode ? `(${errCode})` : ''}: ${msg}` : msg}${detailText}`, type: 'error', duration: 6000 }); @@ -1437,6 +1451,12 @@ export async function initializeLegacySocket(ctx: any) { return; } + if (typeof ctx.appendSystemAction === 'function') { + ctx.appendSystemAction( + `${code ? `[API ${code}] ` : '[API] '}${errType ? `${errType}${errCode ? `(${errCode})` : ''}: ` : ''}${msg}${detailText}` + ); + } + // 最后一次报错：恢复输入状态并清理提示动画 const msgIndex = typeof ctx.currentMessageIndex === 'number' ? ctx.currentMessageIndex : -1; if (msgIndex >= 0 && Array.isArray(ctx.messages)) { diff --git a/utils/api_client.py b/utils/api_client.py index 0ef2f86..44141bd 100644 --- a/utils/api_client.py +++ b/utils/api_client.py @@ -727,13 +727,18 @@ class DeepSeekClient: self.last_error_info = None yield response.json() - except httpx.ConnectError: - self._print(f"{OUTPUT_FORMATS['error']} 无法连接到API服务器，请检查网络连接") + except httpx.ConnectError as e: + connect_detail = str(e).strip() or repr(e) + self._print( + f"{OUTPUT_FORMATS['error']} 无法连接到API服务器，请检查网络连接" + f"（{connect_detail}）" + ) self.last_error_info = { "status_code": None, "error_text": "connect_error", "error_type": "connection_error", - "error_message": "无法连接到API服务器", + "error_message": f"无法连接到API服务器: {connect_detail}", + "error_detail": connect_detail, "request_dump": str(dump_path), "base_url": api_config.get("base_url"), "model_id": api_config.get("model_id"), @@ -744,12 +749,13 @@ class DeepSeekClient: "event": "connect_error", "status_code": None, "error_text": "connect_error", + "error_detail": connect_detail, "base_url": api_config.get("base_url"), "model_id": api_config.get("model_id"), "model_key": self.model_key, "request_dump": str(dump_path) }) - self._mark_request_error(dump_path, error_text="connect_error") + self._mark_request_error(dump_path, error_text=f"connect_error: {connect_detail}") yield {"error": self.last_error_info} except httpx.TimeoutException: self._print(f"{OUTPUT_FORMATS['error']} API请求超时") @@ -776,12 +782,13 @@ class DeepSeekClient: self._mark_request_error(dump_path, error_text="timeout") yield {"error": self.last_error_info} except Exception as e: - self._print(f"{OUTPUT_FORMATS['error']} API调用异常: {e}") + error_text = str(e).strip() or repr(e) + self._print(f"{OUTPUT_FORMATS['error']} API调用异常: {error_text}") self.last_error_info = { "status_code": None, - "error_text": str(e), + "error_text": error_text, "error_type": "exception", - "error_message": str(e), + "error_message": error_text, "request_dump": str(dump_path), "base_url": api_config.get("base_url"), "model_id": api_config.get("model_id"), @@ -791,13 +798,13 @@ class DeepSeekClient: self._debug_log({ "event": "exception", "status_code": None, - "error_text": str(e), + "error_text": error_text, "base_url": api_config.get("base_url"), "model_id": api_config.get("model_id"), "model_key": self.model_key, "request_dump": str(dump_path) }) - self._mark_request_error(dump_path, error_text=str(e)) + self._mark_request_error(dump_path, error_text=error_text) yield {"error": self.last_error_info} async def chat_with_tools(