python 多进程下按天分割日志handler。ConcurrentDayRotatingFileHandler
这个多进程切片安全的python按时间切割文件。
官方的 TimedRotatingFileHandler 在多进程下疯狂报错,
不信的话可以试试官方 TimedRotatingFileHandler 多进程写入文件日志,设置成每秒换一个新的文件写(主要是按天来切割要耽误很长的时间才能观察错误)
此日志handler采用批量聚合每隔1秒写入,在超高速写入时候,写入速度远超官方。
import os import time from pathlib import Path import queue import re import atexit from threading import Lock, Thread from nb_filelock import FileLock import logging # noinspection PyUnresolvedReferences from logging import LogRecord, FileHandler # noinspection PyPep8Naming class ConcurrentDayRotatingFileHandler(logging.Handler): """ 这个多进程切片安全的。 官方的 TimedRotatingFileHandler 在多进程下疯狂报错, 不信的话可以试试官方 TimedRotatingFileHandler 多进程写入文件日志,设置成每秒换一个新的文件写(主要是按天来切割要耽误很长的时间才能观察错误) """ file_handler_list = [] has_start_emit_all_file_handler_process_id_set = set() # 这个linux和windwos都兼容,windwos是多进程每个进程的变量has_start_emit_all_file_handler是独立的。linux是共享的。 __lock_for_rotate = Lock() @classmethod def _emit_all_file_handler(cls): while True: for hr in cls.file_handler_list: # very_nb_print(hr.buffer_msgs_queue.qsize()) # noinspection PyProtectedMember hr._write_to_file() time.sleep(1) # 每隔一秒钟批量写入一次,性能好了很多。 @classmethod def _start_emit_all_file_handler(cls): pass Thread(target=cls._emit_all_file_handler, daemon=True).start() # noinspection PyMissingConstructor def __init__(self, file_name: str, file_path: str, back_count=10): super().__init__() self.file_name = file_name self.file_path = file_path self.backupCount = back_count self.extMatch = re.compile(r"^\d{4}-\d{2}-\d{2}(\.\w+)?$", re.ASCII) self.extMatch2 = re.compile(r"^\d{2}-\d{2}-\d{2}(\.\w+)?$", re.ASCII) self.buffer_msgs_queue = queue.Queue() atexit.register(self._write_to_file) # 如果程序属于立马就能结束的,需要在程序结束前执行这个钩子,防止不到最后一秒的日志没记录到。 self.file_handler_list.append(self) if os.getpid() not in self.has_start_emit_all_file_handler_process_id_set: self._start_emit_all_file_handler() self.__class__.has_start_emit_all_file_handler_process_id_set.add(os.getpid()) def emit(self, record: LogRecord): """ emit已经在logger的handle方法中加了锁,所以这里的重置上次写入时间和清除buffer_msgs不需要加锁了。 :param record: :return: """ # noinspection PyBroadException try: msg = self.format(record) self.buffer_msgs_queue.put(msg) except Exception: self.handleError(record) def _write_to_file(self): buffer_msgs = '' while True: try: msg = self.buffer_msgs_queue.get(block=False) buffer_msgs += msg + '\n' except queue.Empty: break if buffer_msgs: time_str = time.strftime('%Y-%m-%d') # time_str = time.strftime('%H-%M-%S') # 方便测试用的,方便观察。 new_file_name = self.file_name + '.' + time_str path_obj = Path(self.file_path) / Path(new_file_name) path_obj.touch(exist_ok=True) with path_obj.open(mode='a') as f: f.write(buffer_msgs) with FileLock(self.file_path / Path(f'_delete_{self.file_name}.lock')): self._find_and_delete_files() def _find_and_delete_files(self): """ 这一段命名不规范是复制原来的官方旧代码。 Determine the files to delete when rolling over. More specific than the earlier method, which just used glob.glob(). """ dirName = self.file_path baseName = self.file_name fileNames = os.listdir(dirName) result = [] prefix = baseName + "." plen = len(prefix) for fileName in fileNames: if fileName[:plen] == prefix: suffix = fileName[plen:] # print(fileName, prefix,suffix) if self.extMatch.match(suffix) or self.extMatch2.match(suffix): result.append(os.path.join(dirName, fileName)) if len(result) < self.backupCount: result = [] else: result.sort() result = result[:len(result) - self.backupCount] # print(result) for r in result: Path(r).unlink() from concurrent.futures import ProcessPoolExecutor from logging.handlers import TimedRotatingFileHandler logger = logging.getLogger('lala') file_handler = ConcurrentDayRotatingFileHandler('test_my_cd.log', file_path='/pythonlogs') # file_handler = FileHandler('/pythonlogs/test_fhh.log') # 对比官方测试性能 # file_handler = TimedRotatingFileHandler('/pythonlogs/test_fhh.log') # 对比官方测试性能 file_handler.setFormatter(logging.Formatter('[%(asctime)s] - %(filename)s] - %(levelname)s: %(message)s')) logger.addHandler(file_handler) pool = ProcessPoolExecutor(10) def fun(x): for i in range(100000): # time.sleep(0.2) logger.warning(f"{x} {i}") if __name__ == '__main__': print('开始', time.strftime('%H_%M_%S')) for j in range(10): pool.submit(fun, j) pool.shutdown() print('结束', time.strftime('%H_%M_%S'))
反对极端面向过程编程思维方式,喜欢面向对象和设计模式的解读,喜欢对比极端面向过程编程和oop编程消耗代码代码行数的区别和原因。致力于使用oop和36种设计模式写出最高可复用的框架级代码和使用最少的代码行数完成任务,致力于使用oop和设计模式来使部分代码减少90%行,使绝大部分py文件最低减少50%-80%行的写法。