jettask 0.2.18__py3-none-any.whl → 0.2.20__py3-none-any.whl

jettask/executor/executor.py

@@ -0,0 +1,338 @@
+"""
+统一执行器
+
+整合了单进程和多进程执行模式的统一接口
+"""
+
+import asyncio
+import multiprocessing
+import logging
+import os
+import time
+from collections import deque
+from typing import List, Optional
+
+from .core import ExecutionMode, ExecutorCore
+from .orchestrator import ProcessConfig, ProcessOrchestrator
+from ..worker.lifecycle import WorkerStateManager
+from ..utils.rate_limit.manager import RateLimiterManager
+
+logger = logging.getLogger('app')
+
+# Try to use uvloop for better performance
+try:
+    import uvloop
+    uvloop.install()
+    logger.debug("Using uvloop for better performance")
+except ImportError:
+    pass
+
+
+class UnifiedExecutor:
+    """
+    统一执行器
+
+    整合AsyncioExecutor和MultiAsyncioExecutor的功能
+    支持单进程和多进程两种执行模式
+
+    职责:
+    1. 提供统一的执行器接口
+    2. 根据模式选择ExecutorCore或ProcessOrchestrator
+    3. 管理事件队列和任务分发
+    """
+
+    def __init__(self, event_queue, app, concurrency=100,
+                 mode: ExecutionMode = ExecutionMode.SINGLE_PROCESS,
+                 task_name: str = None):
+        """
+        初始化统一执行器
+
+        Args:
+            event_queue: 事件队列
+            app: Application实例
+            concurrency: 并发数
+            mode: 执行模式
+            task_name: 任务名称(单进程模式必需)
+        """
+        self.event_queue = event_queue
+        self.app = app
+        self.concurrency = concurrency
+        self.mode = mode
+        self.task_name = task_name
+
+        # 根据模式初始化核心组件
+        if mode == ExecutionMode.SINGLE_PROCESS:
+            if not task_name:
+                raise ValueError("task_name is required for SINGLE_PROCESS mode")
+
+            self.executor_core = ExecutorCore(
+                app=app,
+                task_name=task_name,
+                concurrency=concurrency
+            )
+            self.orchestrator = None
+            logger.debug(f"UnifiedExecutor initialized in SINGLE_PROCESS mode for task {task_name}")
+
+        elif mode == ExecutionMode.MULTI_PROCESS:
+            self.executor_core = None
+            self.orchestrator = ProcessOrchestrator(
+                app=app,
+                num_processes=concurrency
+            )
+            logger.debug(f"UnifiedExecutor initialized in MULTI_PROCESS mode with {concurrency} processes")
+
+        else:
+            raise ValueError(f"Unsupported execution mode: {mode}")
+
+        # 活动任务集合(单进程模式使用)
+        self._active_tasks = set()
+
+    def logic(self, *args, **kwargs):
+        """
+        BaseExecutor接口方法
+        在单进程模式下不使用,多进程模式委托给ProcessOrchestrator
+        """
+        pass
+
+    async def loop(self):
+        """主循环 - 单进程模式"""
+        if self.mode != ExecutionMode.SINGLE_PROCESS:
+            raise RuntimeError("loop() is only for SINGLE_PROCESS mode")
+
+        # 初始化限流器
+        self.app.consumer_manager._heartbeat_strategy._ensure_consumer_id()
+        worker_id = self.app.consumer_manager._heartbeat_strategy.consumer_id
+        registry_manager = self.app.consumer_manager
+
+        if not self.app.worker_state_manager:
+            self.app.worker_state_manager = WorkerStateManager(
+                redis_client=self.app.ep.async_redis_client,
+                redis_prefix=self.executor_core.prefix,
+                event_pool=self.app.ep  # 传入 EventPool 实例，启用事件驱动的消息恢复
+            )
+            await self.app.worker_state_manager.start_listener()
+            logger.debug(f"WorkerStateManager started for worker {worker_id}")
+
+        # # 初始化时间同步
+        from jettask.utils.time_sync import init_time_sync
+        time_sync = await init_time_sync(self.app.ep.async_redis_client)
+        logger.debug(f"TimeSync initialized, offset={time_sync.get_offset():.6f}s")
+
+        self.executor_core.rate_limiter_manager = RateLimiterManager(
+            redis_client=self.app.ep.async_redis_client,
+            worker_id=worker_id,
+            redis_prefix=self.executor_core.prefix,
+            registry_manager=registry_manager,
+            worker_state_manager=self.app.worker_state_manager
+        )
+        logger.debug(f"RateLimiterManager initialized for worker {worker_id}")
+
+        await self.executor_core.rate_limiter_manager.load_config_from_redis()
+
+        tasks_batch = []
+        max_buffer_size = 5000
+
+        try:
+            while True:
+                # 检查退出信号
+                if hasattr(self.app, '_should_exit') and self.app._should_exit:
+                    logger.debug("UnifiedExecutor detected shutdown signal")
+                    break
+
+                # 检查父进程
+                if hasattr(os, 'getppid') and os.getppid() == 1:
+                    logger.debug("Parent process died, exiting...")
+                    break
+
+                current_time = time.time()
+
+                # 获取事件
+                event = None
+                try:
+                    event = await asyncio.wait_for(self.event_queue.get(), timeout=0.1)
+                except asyncio.TimeoutError:
+                    event = None
+
+                if event:
+                    event.pop("execute_time", None)
+                    tasks_batch.append(event)
+                    logger.debug(f"[EVENT] Got event: {event.get('event_id', 'unknown')}, task_name={event.get('event_data', {}).get('_task_name')}")
+
+                # 批量创建任务
+                if tasks_batch:
+                    for event in tasks_batch:
+                        event_data = event.get('event_data', {})
+                        event_task_name = event_data.get("_task_name") or event_data.get("name")
+
+                        if not event_task_name:
+                            logger.error(f"No task_name in event {event.get('event_id')}")
+                            continue
+
+                        # 验证任务名称匹配
+                        if event_task_name != self.task_name:
+                            logger.error(f"Task name mismatch: {event_task_name} != {self.task_name}")
+                            continue
+
+                        # 限流控制
+                        logger.debug(f"[TASK] Attempting to acquire rate limit for {self.task_name}, event_id={event.get('event_id')}")
+                        rate_limit_token = await self.executor_core.rate_limiter_manager.acquire(
+                            task_name=self.task_name,
+                            timeout=None
+                        )
+                        print(f'{rate_limit_token=}')
+                        if not rate_limit_token:
+                            logger.error(f"Failed to acquire token for {self.task_name}")
+                            continue
+                        logger.debug(f"[TASK] Successfully acquired rate limit for {self.task_name}, token={rate_limit_token}, starting execution")
+
+                        self.executor_core.batch_counter += 1
+
+                        # 创建任务包装器，在任务完成时自动释放限流许可
+                        async def execute_with_release(event_data, token):
+                            try:
+                                await self.executor_core.execute_task(**event_data)
+                            finally:
+                                # 无论任务成功还是失败，都释放并发许可
+                                await self.executor_core.rate_limiter_manager.release(self.task_name, task_id=token)
+
+                        task = asyncio.create_task(execute_with_release(event, rate_limit_token))
+                        self._active_tasks.add(task)
+                        task.add_done_callback(self._active_tasks.discard)
+
+                    tasks_batch.clear()
+
+                # 智能缓冲区管理
+                buffer_full = (
+                    len(self.executor_core.pending_acks) >= max_buffer_size or
+                    len(self.executor_core.status_updates) >= max_buffer_size or
+                    len(self.executor_core.data_updates) >= max_buffer_size or
+                    len(self.executor_core.task_info_updates) >= max_buffer_size
+                )
+
+                should_flush_periodic = False
+                has_pending_data = (
+                    self.executor_core.pending_acks or
+                    self.executor_core.status_updates or
+                    self.executor_core.data_updates or
+                    self.executor_core.task_info_updates
+                )
+
+                if has_pending_data:
+                    for data_type, config in self.executor_core.pipeline_config.items():
+                        time_since_flush = current_time - self.executor_core.last_pipeline_flush[data_type]
+
+                        if data_type == 'ack' and self.executor_core.pending_acks:
+                            if time_since_flush >= config['max_delay']:
+                                should_flush_periodic = True
+                                break
+                        elif data_type == 'task_info' and self.executor_core.task_info_updates:
+                            if time_since_flush >= config['max_delay']:
+                                should_flush_periodic = True
+                                break
+                        elif data_type == 'status' and self.executor_core.status_updates:
+                            if time_since_flush >= config['max_delay']:
+                                should_flush_periodic = True
+                                break
+                        elif data_type == 'data' and self.executor_core.data_updates:
+                            if time_since_flush >= config['max_delay']:
+                                should_flush_periodic = True
+                                break
+
+                if buffer_full or should_flush_periodic:
+                    asyncio.create_task(self.executor_core._flush_all_buffers())
+
+                # 智能休眠
+                has_events = False
+                if isinstance(self.event_queue, deque):
+                    has_events = bool(self.event_queue)
+                elif isinstance(self.event_queue, asyncio.Queue):
+                    has_events = not self.event_queue.empty()
+
+                if has_events:
+                    await asyncio.sleep(0)
+                else:
+                    if has_pending_data:
+                        await self.executor_core._flush_all_buffers()
+                    await asyncio.sleep(0.001)
+
+        except KeyboardInterrupt:
+            logger.debug("UnifiedExecutor received KeyboardInterrupt")
+        except Exception as e:
+            logger.error(f"UnifiedExecutor loop error: {e}")
+        finally:
+            await self._cleanup_single_process()
+
+    async def _cleanup_single_process(self):
+        """清理单进程模式资源"""
+        logger.debug("UnifiedExecutor cleaning up...")
+
+        # 设置停止标志
+        if hasattr(self.app.ep, '_stop_reading'):
+            self.app.ep._stop_reading = True
+
+        # 取消活动任务
+        if self._active_tasks:
+            logger.debug(f"Cancelling {len(self._active_tasks)} active tasks...")
+            for task in self._active_tasks:
+                if not task.done():
+                    task.cancel()
+
+            if self._active_tasks:
+                try:
+                    await asyncio.wait_for(
+                        asyncio.gather(*self._active_tasks, return_exceptions=True),
+                        timeout=0.2
+                    )
+                except asyncio.TimeoutError:
+                    logger.debug("Some tasks did not complete in time")
+
+        # 清理ExecutorCore
+        await self.executor_core.cleanup()
+
+        # 清理event_pool
+        if hasattr(self.app.ep, 'cleanup'):
+            try:
+                self.app.ep.cleanup()
+            except Exception as e:
+                logger.error(f"Error cleaning up EventPool: {e}")
+
+        # 标记worker离线
+        if self.app.consumer_manager:
+            try:
+                self.app.consumer_manager.cleanup()
+                logger.debug("Worker marked as offline")
+            except Exception as e:
+                logger.error(f"Error marking worker offline: {e}")
+
+        logger.debug("UnifiedExecutor stopped")
+
+    def start_multi_process(self, queues: List[str], prefetch_multiplier: int = 100, worker_id: str = None, worker_key: str = None):
+        """启动多进程模式
+
+        Args:
+            queues: 队列列表
+            prefetch_multiplier: 预取倍数
+            worker_id: Worker ID（主进程生成，子进程复用）
+            worker_key: Worker Key（主进程生成，子进程复用）
+        """
+        if self.mode != ExecutionMode.MULTI_PROCESS:
+            raise RuntimeError("start_multi_process() is only for MULTI_PROCESS mode")
+
+        self.orchestrator.start(queues, prefetch_multiplier, worker_id, worker_key)
+
+    def shutdown(self):
+        """
+        关闭执行器
+
+        根据执行模式调用相应的关闭方法
+        """
+        if self.mode == ExecutionMode.MULTI_PROCESS:
+            if self.orchestrator:
+                self.orchestrator.shutdown()
+        elif self.mode == ExecutionMode.SINGLE_PROCESS:
+            # 单进程模式的清理在 _cleanup_single_process 中处理
+            # 这里只是一个占位符，实际清理由事件循环完成
+            logger.debug("UnifiedExecutor shutdown called in SINGLE_PROCESS mode")
+
+
+__all__ = ['UnifiedExecutor']

jettask/executor/orchestrator.py

@@ -0,0 +1,290 @@
+"""
+多进程编排器
+
+职责：
+1. 启动和停止子进程
+2. 监控子进程健康
+3. 故障自动重启
+
+注意：使用系统默认的 multiprocessing 启动模式（Linux上是fork，Windows/macOS上是spawn）。
+- fork模式：子进程会继承父进程的状态，需要通过 cleanup_inherited_state() 清理
+- spawn模式：子进程是全新的Python解释器，不会继承父进程状态，但需要所有对象可序列化
+"""
+import time
+import os
+import signal
+import logging
+import multiprocessing
+from dataclasses import dataclass
+from typing import Dict, List
+
+logger = logging.getLogger('app')
+
+
+@dataclass
+class ProcessConfig:
+    """进程配置"""
+    process_id: int
+    redis_url: str
+    redis_prefix: str
+    queues: List[str]
+    tasks: Dict
+    concurrency: int
+    prefetch_multiplier: int
+    max_connections: int
+    consumer_strategy: str
+    consumer_config: Dict
+    worker_id: str
+    worker_key: str
+
+
+class ProcessOrchestrator:
+    """
+    进程编排器 - 只负责进程生命周期管理
+
+    职责：
+    1. 启动和停止子进程
+    2. 监控子进程健康状态
+    3. 自动重启失败的进程
+    """
+
+    def __init__(self, app, num_processes: int = 2):
+        """
+        初始化进程编排器
+
+        Args:
+            app: Application 实例
+            num_processes: 进程数量
+        """
+        self.app = app
+        self.num_processes = num_processes
+        self.processes: Dict[int, multiprocessing.Process] = {}
+        self.process_configs: Dict[int, ProcessConfig] = {}
+        self.shutdown_event = multiprocessing.Event()
+
+        # 监控配置
+        self._monitor_interval = 1.0
+        self._restart_delay = 2.0
+        self._max_restart_attempts = 3
+        self._restart_counts: Dict[int, int] = {}
+        self._main_received_signal = False
+        self._shutdown_called = False
+
+        logger.debug(f"ProcessOrchestrator initialized with {num_processes} processes")
+
+    def _create_process_config(
+        self,
+        process_id: int,
+        queues: List[str],
+        prefetch_multiplier: int
+    ) -> ProcessConfig:
+        """创建进程配置"""
+        # 复制 consumer_config 并添加 disable_heartbeat_process 标志
+        # 因为在子进程中，心跳已经由主进程的 HeartbeatThreadManager 管理
+        consumer_config = dict(self.app.consumer_config or {})
+        consumer_config['disable_heartbeat_process'] = True
+
+        return ProcessConfig(
+            process_id=process_id,
+            redis_url=self.app.redis_url,
+            redis_prefix=self.app.redis_prefix,
+            queues=queues,
+            tasks=self.app._tasks,
+            concurrency=10000,  # 子进程内部并发
+            prefetch_multiplier=prefetch_multiplier,
+            max_connections=self.app.max_connections,
+            consumer_strategy=self.app.consumer_strategy,
+            consumer_config=consumer_config,
+            worker_id=getattr(self.app, 'worker_id', None),
+            worker_key=getattr(self.app, 'worker_key', None)
+        )
+
+    def _start_process(self, process_id: int, config: ProcessConfig) -> multiprocessing.Process:
+        """启动单个子进程
+
+        注意：在 fork 模式下，子进程会在 subprocess_main 的第一行重置 logging 锁。
+        """
+        from .process_entry import subprocess_main
+        process = multiprocessing.Process(
+            target=subprocess_main,
+            args=(
+                config.process_id,
+                config.redis_url,
+                config.redis_prefix,
+                config.queues,
+                config.tasks,
+                config.concurrency,
+                config.prefetch_multiplier,
+                config.max_connections,
+                config.consumer_strategy,
+                config.consumer_config,
+                config.worker_id,
+                config.worker_key,
+                self.shutdown_event
+            ),
+            name=f"JetTask-Worker-{process_id}"
+        )
+        process.start()
+
+
+        logger.info(f"Started process #{process_id} (PID: {process.pid})")
+        return process
+
+    def _restart_process(self, process_id: int):
+        """重启失败的进程"""
+        if self.shutdown_event.is_set():
+            return
+
+        restart_count = self._restart_counts.get(process_id, 0)
+        if restart_count >= self._max_restart_attempts:
+            logger.error(f"Process #{process_id} exceeded max restart attempts")
+            return
+
+        self._restart_counts[process_id] = restart_count + 1
+        delay = self._restart_delay * (2 ** restart_count)
+
+        logger.info(
+            f"Restarting process #{process_id} (attempt {restart_count + 1}) after {delay}s"
+        )
+        time.sleep(delay)
+
+        config = self.process_configs[process_id]
+        process = self._start_process(process_id, config)
+        self.processes[process_id] = process
+
+    def _monitor_processes(self) -> int:
+        """监控进程健康状态"""
+        alive_count = 0
+
+        for process_id, process in list(self.processes.items()):
+            if process.is_alive():
+                alive_count += 1
+                self._restart_counts[process_id] = 0  # 重置重启计数
+            else:
+                exit_code = process.exitcode
+
+                if self.shutdown_event.is_set():
+                    logger.debug(f"Process #{process_id} stopped during shutdown")
+                elif exit_code in (-15, -2):  # SIGTERM, SIGINT
+                    logger.info(f"Process #{process_id} received termination signal")
+                    self.shutdown_event.set()
+                elif exit_code == 0:
+                    if not self._main_received_signal:
+                        # 检查是否所有进程都停止了
+                        all_stopped = all(not p.is_alive() for p in self.processes.values())
+                        if all_stopped:
+                            logger.info("All processes stopped simultaneously")
+                            self.shutdown_event.set()
+                        else:
+                            logger.warning(f"Process #{process_id} stopped unexpectedly")
+                            self._restart_process(process_id)
+                else:
+                    logger.error(f"Process #{process_id} exited with code {exit_code}")
+                    self._restart_process(process_id)
+
+        return alive_count
+
+    def start(
+        self,
+        queues: List[str],
+        prefetch_multiplier: int = 100,
+        worker_ids: list = None,
+    ):
+        """启动所有进程
+
+        Args:
+            queues: 队列列表
+            prefetch_multiplier: 预取倍数
+        """
+        logger.info(f"Starting {self.num_processes} worker processes")
+
+        # 设置主进程信号处理
+        def signal_handler(signum, frame):
+            logger.info(f"Main process received signal {signum}")
+            self._main_received_signal = True
+            self.shutdown_event.set()
+
+        signal.signal(signal.SIGINT, signal_handler)
+        signal.signal(signal.SIGTERM, signal_handler)
+
+        try:
+            # 启动所有进程
+            for i, (worker_id, worker_key) in enumerate(worker_ids):
+                config = self._create_process_config(i, queues, prefetch_multiplier)
+                if worker_id:
+                    config.worker_id = worker_id
+                    config.worker_key = worker_key
+
+                self.process_configs[i] = config
+                process = self._start_process(i, config)
+                self.processes[i] = process
+
+                time.sleep(0.1)  # 错开启动时间
+
+            logger.info(f"All {self.num_processes} processes started")
+
+            # 监控循环
+            while not self.shutdown_event.is_set():
+                alive_count = self._monitor_processes()
+
+                if alive_count == 0:
+                    if self._main_received_signal or self.shutdown_event.is_set():
+                        logger.info("All processes stopped during shutdown")
+                    else:
+                        logger.error("All processes stopped unexpectedly")
+                    break
+
+                time.sleep(self._monitor_interval)
+
+        except KeyboardInterrupt:
+            logger.info("Main process received KeyboardInterrupt")
+            self._main_received_signal = True
+            self.shutdown_event.set()
+        except Exception as e:
+            logger.error(f"ProcessOrchestrator error: {e}", exc_info=True)
+        finally:
+            self.shutdown()
+
+    def shutdown(self):
+        """关闭所有进程"""
+        if self._shutdown_called:
+            return
+        self._shutdown_called = True
+
+        logger.info("Shutting down ProcessOrchestrator")
+
+        self.shutdown_event.set()
+
+        # 强制杀死所有进程
+        for process_id, process in self.processes.items():
+            if process.is_alive():
+                logger.info(f"Force killing process #{process_id} (PID: {process.pid})")
+                try:
+                    os.kill(process.pid, signal.SIGKILL)
+                except Exception as e:
+                    logger.warning(f"Error killing process #{process_id}: {e}")
+                    try:
+                        process.kill()
+                    except:
+                        pass
+
+        # 等待进程退出（最多1秒）
+        start_time = time.time()
+        for process_id, process in self.processes.items():
+            remaining = max(0, 1.0 - (time.time() - start_time))
+            try:
+                process.join(timeout=remaining)
+            except:
+                pass
+
+            if process.is_alive():
+                logger.warning(f"Process #{process_id} still alive after SIGKILL")
+
+        self.processes.clear()
+        self.process_configs.clear()
+        self._restart_counts.clear()
+
+        logger.info("ProcessOrchestrator shutdown complete")
+
+
+__all__ = ['ProcessOrchestrator', 'ProcessConfig']