task_server.py 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432
  1. import json
  2. import threading
  3. import time
  4. from concurrent.futures import ThreadPoolExecutor
  5. from datetime import datetime
  6. from typing import Dict
  7. from sqlalchemy import func
  8. from pqai_agent import logging_service
  9. from pqai_agent.agents.message_push_agent import MessagePushAgent
  10. from pqai_agent.agents.multimodal_chat_agent import MultiModalChatAgent
  11. from pqai_agent.data_models.agent_configuration import AgentConfiguration
  12. from pqai_agent.data_models.agent_test_task import AgentTestTask
  13. from pqai_agent.data_models.agent_test_task_conversations import AgentTestTaskConversations
  14. from pqai_agent.data_models.service_module import ServiceModule
  15. from pqai_agent_server.const.status_enum import TestTaskConversationsStatus, TestTaskStatus, get_test_task_status_desc
  16. logger = logging_service.logger
  17. class TaskManager:
  18. """任务管理器"""
  19. def __init__(self, session_maker, dataset_service):
  20. self.session_maker = session_maker
  21. self.dataset_service = dataset_service
  22. self.task_events = {} # 任务ID -> Event (用于取消任务)
  23. self.task_locks = {} # 任务ID -> Lock (用于任务状态同步)
  24. self.running_tasks = set()
  25. self.executor = ThreadPoolExecutor(max_workers=20, thread_name_prefix='TaskWorker')
  26. self.create_task_executor = ThreadPoolExecutor(max_workers=10, thread_name_prefix='CreateTaskWorker')
  27. self.task_futures = {} # 任务ID -> Future
  28. def get_test_task_list(self, page_num: int, page_size: int) -> Dict:
  29. with self.session_maker() as session:
  30. # 计算偏移量
  31. offset = (page_num - 1) * page_size
  32. # 查询分页数据
  33. result = (session.query(AgentTestTask, AgentConfiguration)
  34. .outerjoin(AgentConfiguration, AgentTestTask.agent_id == AgentConfiguration.id)
  35. .limit(page_size).offset(offset).all())
  36. # 查询总记录数
  37. total = session.query(func.count(AgentTestTask.id)).scalar()
  38. total_page = total // page_size + 1 if total % page_size > 0 else total // page_size
  39. total_page = 1 if total_page <= 0 else total_page
  40. response_data = [
  41. {
  42. "id": agent_test_task.id,
  43. "agentName": agent_configuration.name,
  44. "createUser": agent_test_task.create_user,
  45. "updateUser": agent_test_task.update_user,
  46. "statusName": get_test_task_status_desc(agent_test_task.status),
  47. "createTime": agent_test_task.create_time.strftime("%Y-%m-%d %H:%M:%S"),
  48. "updateTime": agent_test_task.update_time.strftime("%Y-%m-%d %H:%M:%S")
  49. }
  50. for agent_test_task, agent_configuration in result
  51. ]
  52. return {
  53. "currentPage": page_num,
  54. "pageSize": page_size,
  55. "totalSize": total_page,
  56. "total": total,
  57. "list": response_data,
  58. }
  59. def get_test_task_conversations(self, task_id: int, page_num: int, page_size: int) -> Dict:
  60. with self.session_maker() as session:
  61. # 计算偏移量
  62. offset = (page_num - 1) * page_size
  63. # 查询分页数据
  64. result = (session.query(AgentTestTaskConversations, AgentConfiguration)
  65. .outerjoin(AgentConfiguration, AgentTestTaskConversations.agent_id == AgentConfiguration.id)
  66. .filter(AgentTestTaskConversations.task_id == task_id)
  67. .limit(page_size).offset(offset).all())
  68. # 查询总记录数
  69. total = session.query(func.count(AgentTestTaskConversations.id)).scalar()
  70. total_page = total // page_size + 1 if total % page_size > 0 else total // page_size
  71. total_page = 1 if total_page <= 0 else total_page
  72. response_data = [
  73. {
  74. "id": agent_test_task_conversation.id,
  75. "agentName": agent_configuration.name,
  76. "input":MultiModalChatAgent.compose_dialogue(json.loads(agent_test_task_conversation.input)),
  77. "output": agent_test_task_conversation.output,
  78. "score": agent_test_task_conversation.score,
  79. "statusName": get_test_task_status_desc(agent_test_task_conversation.status),
  80. "createTime": agent_test_task_conversation.create_time.strftime("%Y-%m-%d %H:%M:%S"),
  81. "updateTime": agent_test_task_conversation.update_time.strftime("%Y-%m-%d %H:%M:%S")
  82. }
  83. for agent_test_task_conversation, agent_configuration in result
  84. ]
  85. return {
  86. "currentPage": page_num,
  87. "pageSize": page_size,
  88. "totalSize": total_page,
  89. "total": total,
  90. "list": response_data,
  91. }
  92. def create_task(self, agent_id: int, module_id: int) -> Dict:
  93. """创建新任务"""
  94. with self.session_maker() as session:
  95. agent_test_task = AgentTestTask(agent_id=agent_id, module_id=module_id,
  96. status=TestTaskStatus.CREATING.value)
  97. session.add(agent_test_task)
  98. session.commit() # 显式提交
  99. task_id = agent_test_task.id
  100. # 异步执行创建任务
  101. self.create_task_executor.submit(self._generate_agent_test_task_conversation_batch, task_id, agent_id,
  102. module_id)
  103. return self.get_task(task_id)
  104. def _generate_agent_test_task_conversation_batch(self, task_id: int, agent_id: int, module_id: int):
  105. """异步生成子任务"""
  106. try:
  107. # 获取数据集列表
  108. dataset_module_list = self.dataset_service.get_dataset_module_list_by_module(module_id)
  109. # 批量处理数据集 - 减少数据库交互
  110. batch_size = 100 # 每批处理100个子任务
  111. agent_test_task_conversation_batch = []
  112. for dataset_module in dataset_module_list:
  113. # 获取对话数据列表
  114. conversation_datas = self.dataset_service.get_conversation_data_list_by_dataset(dataset_module.dataset_id)
  115. for conversation_data in conversation_datas:
  116. # 创建子任务对象
  117. agent_test_task_conversation = AgentTestTaskConversations(
  118. task_id=task_id,
  119. agent_id=agent_id,
  120. dataset_id=dataset_module.dataset_id,
  121. conversation_id=conversation_data.id,
  122. status=TestTaskConversationsStatus.PENDING.value
  123. )
  124. agent_test_task_conversation_batch.append(agent_test_task_conversation)
  125. # 批量提交
  126. if len(agent_test_task_conversation_batch) >= batch_size:
  127. self.save_agent_test_task_conversation_batch(agent_test_task_conversation_batch)
  128. agent_test_task_conversation_batch = []
  129. # 提交剩余的子任务
  130. if agent_test_task_conversation_batch:
  131. self.save_agent_test_task_conversation_batch(agent_test_task_conversation_batch)
  132. # 更新主任务状态为未开始
  133. self.update_task_status(task_id, TestTaskStatus.NOT_STARTED.value)
  134. # 自动提交任务执行
  135. self._execute_task(task_id)
  136. except Exception as e:
  137. logger.error(f"生成子任务失败: {str(e)}")
  138. # 更新任务状态为失败
  139. self.update_task_status(task_id, TestTaskStatus.CREATED_FAIL.value)
  140. def save_agent_test_task_conversation_batch(self, agent_test_task_conversation_batch: list):
  141. """批量保存子任务到数据库"""
  142. try:
  143. with self.session_maker() as session:
  144. with session.begin():
  145. session.add_all(agent_test_task_conversation_batch)
  146. except Exception as e:
  147. logger.error(e)
  148. def get_agent_configuration_by_task_id(self, task_id: int):
  149. """获取指定任务ID对应的Agent配置信息"""
  150. with self.session_maker() as session:
  151. return session.query(AgentConfiguration) \
  152. .join(AgentTestTask, AgentTestTask.agent_id == AgentConfiguration.id) \
  153. .filter(AgentTestTask.id == task_id) \
  154. .one_or_none() # 返回单个对象或None(如果未找到)
  155. def get_service_module_by_task_id(self, task_id: int):
  156. """获取指定任务ID对应的Agent配置信息"""
  157. with self.session_maker() as session:
  158. return session.query(ServiceModule) \
  159. .join(AgentTestTask, AgentTestTask.module_id == ServiceModule.id) \
  160. .filter(AgentTestTask.id == task_id) \
  161. .one_or_none() # 返回单个对象或None(如果未找到)
  162. def get_task(self, task_id: int):
  163. """获取任务信息"""
  164. with self.session_maker() as session:
  165. return session.query(AgentTestTask).filter(AgentTestTask.id == task_id).one()
  166. def get_in_progress_task(self):
  167. """获取执行中任务"""
  168. with self.session_maker() as session:
  169. return session.query(AgentTestTask).filter(AgentTestTask.status == TestTaskStatus.IN_PROGRESS.value).all()
  170. def get_creating_task(self):
  171. """获取执行中任务"""
  172. with self.session_maker() as session:
  173. return session.query(AgentTestTask).filter(AgentTestTask.status == TestTaskStatus.CREATING.value).all()
  174. def get_task_conversations(self, task_id: int):
  175. """获取任务的所有子任务"""
  176. with self.session_maker() as session:
  177. return session.query(AgentTestTaskConversations).filter(AgentTestTaskConversations.task_id == task_id).all()
  178. def del_task_conversations(self, task_id: int):
  179. with self.session_maker() as session:
  180. session.query(AgentTestTaskConversations).filter(AgentTestTaskConversations.task_id == task_id).delete()
  181. # 提交事务生效
  182. session.commit()
  183. def get_pending_task_conversations(self, task_id: int):
  184. """获取待处理的子任务"""
  185. with self.session_maker() as session:
  186. return session.query(AgentTestTaskConversations).filter(
  187. AgentTestTaskConversations.task_id == task_id).filter(
  188. AgentTestTaskConversations.status.in_([
  189. TestTaskConversationsStatus.PENDING.value,
  190. TestTaskConversationsStatus.RUNNING.value
  191. ])).all()
  192. def update_task_status(self, task_id: int, status: int):
  193. """更新任务状态"""
  194. with self.session_maker() as session:
  195. session.query(AgentTestTask).filter(AgentTestTask.id == task_id).update(
  196. {"status": status, "update_time": datetime.now()})
  197. session.commit()
  198. def update_task_conversations_status(self, task_conversations_id: int, status: int):
  199. """更新子任务状态"""
  200. with self.session_maker() as session:
  201. session.query(AgentTestTaskConversations).filter(
  202. AgentTestTaskConversations.id == task_conversations_id).update(
  203. {"status": status, "update_time": datetime.now()})
  204. session.commit()
  205. def update_task_conversations_res(self, task_conversations_id: int, status: int, input: str, output: str,
  206. score: str):
  207. """更新子任务结果"""
  208. with self.session_maker() as session:
  209. session.query(AgentTestTaskConversations).filter(
  210. AgentTestTaskConversations.id == task_conversations_id).update(
  211. {"status": status, "input": input, "output": output, "score": score, "update_time": datetime.now()})
  212. session.commit()
  213. def cancel_task(self, task_id: int):
  214. """取消任务(带事务支持)"""
  215. # 设置取消事件
  216. if task_id in self.task_events:
  217. self.task_events[task_id].set()
  218. # 如果任务正在执行,尝试取消Future
  219. if task_id in self.task_futures:
  220. self.task_futures[task_id].cancel()
  221. with self.session_maker() as session:
  222. with session.begin():
  223. session.query(AgentTestTask).filter(AgentTestTask.id == task_id).update(
  224. {"status": TestTaskStatus.CANCELLED.value})
  225. session.query(AgentTestTaskConversations).filter(AgentTestTaskConversations.task_id == task_id).filter(
  226. AgentTestTaskConversations.status == TestTaskConversationsStatus.PENDING.value).update(
  227. {"status": TestTaskConversationsStatus.CANCELLED.value})
  228. session.commit()
  229. def resume_task(self, task_id: int) -> bool:
  230. """恢复已取消的任务"""
  231. task = self.get_task(task_id)
  232. if not task or task.status != TestTaskStatus.CANCELLED.value:
  233. return False
  234. with self.session_maker() as session:
  235. with session.begin():
  236. session.query(AgentTestTask).filter(AgentTestTask.id == task_id).update(
  237. {"status": TestTaskStatus.NOT_STARTED.value})
  238. session.query(AgentTestTaskConversations).filter(AgentTestTaskConversations.task_id == task_id).filter(
  239. AgentTestTaskConversations.status == TestTaskConversationsStatus.CANCELLED.value).update(
  240. {"status": TestTaskConversationsStatus.PENDING.value})
  241. session.commit()
  242. # 重新执行任务
  243. self._execute_task(task_id)
  244. logger.info(f"Resumed task {task_id}")
  245. return True
  246. def recover_tasks(self):
  247. """服务启动时恢复未完成的任务"""
  248. creating = self.get_creating_task()
  249. for task in creating:
  250. task_id = task.id
  251. agent_id = task.agent_id
  252. module_id = task.module_id
  253. self.del_task_conversations(task_id)
  254. # 重新提交任务
  255. # 异步执行创建任务
  256. self.create_task_executor.submit(self._generate_agent_test_task_conversation_batch, task_id, agent_id,
  257. module_id)
  258. # 获取所有进行中的任务ID(根据实际状态定义查询)
  259. in_progress_tasks = self.get_in_progress_task()
  260. for task in in_progress_tasks:
  261. task_id = task.id
  262. # 重新提交任务
  263. self._execute_task(task_id)
  264. def _execute_task(self, task_id: int):
  265. """提交任务到线程池执行"""
  266. # 确保任务状态一致性
  267. if task_id in self.running_tasks:
  268. return
  269. # 创建任务事件和锁
  270. if task_id not in self.task_events:
  271. self.task_events[task_id] = threading.Event()
  272. if task_id not in self.task_locks:
  273. self.task_locks[task_id] = threading.Lock()
  274. # 提交到线程池
  275. future = self.executor.submit(self._process_task, task_id)
  276. self.task_futures[task_id] = future
  277. # 标记任务为运行中
  278. with self.task_locks[task_id]:
  279. self.running_tasks.add(task_id)
  280. def _process_task(self, task_id: int):
  281. """处理任务的所有子任务"""
  282. try:
  283. # 更新任务状态为运行中
  284. self.update_task_status(task_id, TestTaskStatus.IN_PROGRESS.value)
  285. # 获取所有待处理的子任务
  286. task_conversations = self.get_pending_task_conversations(task_id)
  287. agent_configuration = self.get_agent_configuration_by_task_id(task_id)
  288. query_prompt_template = agent_configuration.task_prompt
  289. agent = MultiModalChatAgent(model=agent_configuration.execution_model,
  290. system_prompt=agent_configuration.system_prompt,
  291. tools=json.loads(agent_configuration.tools))
  292. # 执行每个子任务
  293. for task_conversation in task_conversations:
  294. # 检查任务是否被取消
  295. if self.task_events[task_id].is_set():
  296. break
  297. # 更新子任务状态为运行中
  298. if task_conversation.status == TestTaskConversationsStatus.PENDING.value:
  299. self.update_task_conversations_status(task_conversation.id,
  300. TestTaskConversationsStatus.RUNNING.value)
  301. try:
  302. conversation_data = self.dataset_service.get_conversation_data_by_id(
  303. task_conversation.conversation_id)
  304. user_profile_data = self.dataset_service.get_user_profile_data(conversation_data.user_id,
  305. conversation_data.version_date.replace(
  306. "-", ""))
  307. user_profile = json.loads(user_profile_data['profile_data_v1'])
  308. avatar = user_profile_data['iconurl']
  309. staff_profile_data = self.dataset_service.get_staff_profile_data(
  310. conversation_data.staff_id).agent_profile
  311. conversations = self.dataset_service.get_chat_conversation_list_by_ids(
  312. json.loads(conversation_data.conversation), conversation_data.staff_id)
  313. conversations = sorted(conversations, key=lambda i: i['timestamp'], reverse=False)
  314. last_timestamp = int(conversations[-1]["timestamp"])
  315. push_time = int(last_timestamp / 1000) + 24 * 3600
  316. push_dt = datetime.fromtimestamp(push_time).strftime('%Y-%m-%d %H:%M:%S')
  317. push_message = agent._generate_message(
  318. context={
  319. "formatted_staff_profile": staff_profile_data,
  320. "nickname": user_profile['nickname'],
  321. "name": user_profile['name'],
  322. "avatar": avatar,
  323. "preferred_nickname": user_profile['preferred_nickname'],
  324. "gender": user_profile['gender'],
  325. "age": user_profile['age'],
  326. "region": user_profile['region'],
  327. "health_conditions": user_profile['health_conditions'],
  328. "medications": user_profile['medications'],
  329. "interests": user_profile['interests'],
  330. "current_datetime": push_dt
  331. },
  332. dialogue_history=conversations,
  333. query_prompt_template=query_prompt_template
  334. )
  335. # TODO 获取打分
  336. score = '{"score":0.05}'
  337. # 更新子任务状态为已完成
  338. self.update_task_conversations_res(task_conversation.id,
  339. TestTaskConversationsStatus.SUCCESS.value,
  340. json.dumps(conversations, ensure_ascii=False),
  341. json.dumps(push_message, ensure_ascii=False),
  342. score)
  343. except Exception as e:
  344. logger.error(f"Error executing task {task_id}: {str(e)}")
  345. self.update_task_conversations_status(task_conversation.id,
  346. TestTaskConversationsStatus.FAILED.value)
  347. # 检查任务是否完成
  348. task_conversations = self.get_task_conversations(task_id)
  349. all_completed = all(task_conversation.status in
  350. (TestTaskConversationsStatus.SUCCESS.value, TestTaskConversationsStatus.FAILED.value)
  351. for task_conversation in task_conversations)
  352. any_pending = any(task_conversation.status in
  353. (TestTaskConversationsStatus.PENDING.value, TestTaskConversationsStatus.RUNNING.value)
  354. for task_conversation in task_conversations)
  355. if all_completed:
  356. self.update_task_status(task_id, TestTaskStatus.COMPLETED.value)
  357. logger.info(f"Task {task_id} completed")
  358. elif not any_pending:
  359. # 没有待处理子任务但未全部完成(可能是取消了)
  360. current_status = self.get_task(task_id).status
  361. if current_status != TestTaskStatus.CANCELLED.value:
  362. self.update_task_status(task_id, TestTaskStatus.COMPLETED.value
  363. if all_completed else TestTaskStatus.CANCELLED.value)
  364. except Exception as e:
  365. logger.error(f"Error executing task {task_id}: {str(e)}")
  366. self.update_task_status(task_id, TestTaskStatus.FAILED.value)
  367. finally:
  368. # 清理资源
  369. with self.task_locks[task_id]:
  370. if task_id in self.running_tasks:
  371. self.running_tasks.remove(task_id)
  372. if task_id in self.task_events:
  373. del self.task_events[task_id]
  374. if task_id in self.task_futures:
  375. del self.task_futures[task_id]
  376. def shutdown(self):
  377. """关闭执行器"""
  378. self.executor.shutdown(wait=False)
  379. logger.info("Task executor shutdown")