单线程多定时任务
前言:公司业务需求,实例当中大量需要启动定时器的操作;大家都知道python中的定时器用的是threading.Timer,每当启动一个定时器时,程序内部起了一个线程,定时器触发执行结束后,线程自动销毁;这里就涉及到一个问题,如果同时有大量启动定时器的需求时,内部线程过多,程序肯定就崩了,有没有启一个线程就能完成定时器的操作呢?网上查了一些资料,还没有看到能解决目前问题的现成代码,不如自己搞一个试试
1、初始版本:
思路:定时器,说白了就是延时执行指定的程序,目前自己重构python里面的定时器不太现实,能力达不到,所以延时操作时还得用到系统定时器,不过我们可以改一下规则;把所有要进行定时操作的程序添加到特定列表中,把列表中定时时间最短程序拿出来,进行threading.Timer(time,callback)绑定,等时间超时触发自定义的callback,执行刚刚列表取出的程序;然后把时间更新,再次把列表中时间最短的程序拿出了,继续threading.Timer绑定,不断的迭代循环;当有新的定时任务加入到列表时,把当前的threading.Timer绑定取消,更新列表中的时间,再次取出最短时间,进行threading.Timer绑定......
代码:
import threading
import time
class Timer():
'''单线程下的定时器'''
def __init__(self):
self.queues = []
self.timer = None
self.last_time = time.time()
def start(self):
item = self.get()
if item:
self.timer = threading.Timer(item[0],self.execute)
self.timer.start()
def add(self,item):
print('add',item)
self.flush_time()
self.queues.append(item)
self.queues.sort(key=lambda x:x[0])
if self.timer:
self.timer.cancel()
self.timer = None
self.start()
def get(self):
item = None
if len(self.queues) > 0:
item = self.queues[0]
return item
def pop(self):
item = None
if len(self.queues) > 0:
item = self.queues.pop(0)
return item
def flush_time(self):
curr_time = time.time()
for i in self.queues:
i[0] = i[0] - (curr_time - self.last_time)
self.last_time = curr_time
def execute(self):
# if self.timer:
# self.timer.cancel()
# self.timer = None
item = self.pop()
self.flush_time()
if item:
callback = item[1]
args = item[0]
callback(args)
self.start()
执行及输出:
if __name__ == '__main__': # 检测线程数 def func(): while True: print(threading.active_count()) time.sleep(1) f1 = threading.Thread(target=func) f1.start() import logging logging.basicConfig(level=logging.INFO,format="%(asctime)s %(message)s", datefmt="%m/%d/%Y %H:%M:%S [%A]") def func1(*args): logging.info('func1 %s'%args) # time.sleep(5) def func2(*args): logging.info('func2 %s' % args) # time.sleep(5) def func3(*args): logging.info('func3 %s' % args) # time.sleep(5) def func4(*args): logging.info('func4 %s' % args) # time.sleep(5) def func5(*args): logging.info('func5 %s' % args) # time.sleep(5) # 测试 t1 = Timer() logging.info('start') t1.add([5,func1]) time.sleep(0.5) t1.add([4,func2]) time.sleep(0.5) t1.add([3,func3]) time.sleep(0.5) t1.add([2,func4]) time.sleep(0.5) t1.add([1,func5]) time.sleep(5) t1.add([1,func1]) t1.add([2,func2]) t1.add([3,func3]) t1.add([4,func4]) t1.add([5,func5]) # 输出 # 2 # 07/27/2017 10:36:47 [Thursday] start # add [5, <function func1 at 0x000000D79FC77E18>] # add [4, <function func2 at 0x000000D79FCA8488>] # 3 # add [3, <function func3 at 0x000000D79FCA8510>] # add [2, <function func4 at 0x000000D79FCA8598>] # 3 # add [1, <function func5 at 0x000000D79FCA8620>] # 3 # 07/27/2017 10:36:50 [Thursday] func5 1 # 07/27/2017 10:36:51 [Thursday] func4 0.498349666595459 # 3 # 07/27/2017 10:36:51 [Thursday] func3 0.49782633781433105 # 07/27/2017 10:36:52 [Thursday] func2 0.49848270416259766 # 3 # 07/27/2017 10:36:52 [Thursday] func1 0.48449039459228516 # 2 # 2 # add [1, <function func1 at 0x000000D79FC77E18>] # add [2, <function func2 at 0x000000D79FCA8488>] # add [3, <function func3 at 0x000000D79FCA8510>] # add [4, <function func4 at 0x000000D79FCA8598>] # add [5, <function func5 at 0x000000D79FCA8620>] # 3 # 07/27/2017 10:36:55 [Thursday] func1 0.9990766048431396 # 3 # 07/27/2017 10:36:56 [Thursday] func2 0.9988017082214355 # 3 # 07/27/2017 10:36:57 [Thursday] func3 0.99928879737854 # 07/27/2017 10:36:58 [Thursday] func4 0.9991350173950195 # 3 # 3 # 07/27/2017 10:36:59 [Thursday] func5 0.9988160133361816
注:查看代码输出,所有的定时器都按照标定的时间依次执行,非常完美,一切看起来很美好,只是看起来,呵呵哒,当你把func里面的time.sleep(5)启用后,线程数蹭蹭的上来了;原因是上个定时器callback还是执行中,下个定时器已经启动了,这时就又新增了一个线程,哎,失败
2、修订版本
思路:利用生成者消费者模型,用到threading.Condition条件变量;强制永远启用的是一个Timer!
代码:
import time
import threading
import logging
class NewTimer(threading.Thread):
'''单线程下的定时器'''
def __init__(self):
super().__init__()
self.queues = []
self.timer = None
self.cond = threading.Condition()
def run(self):
while True:
# print('NewTimer',self.queues)
self.cond.acquire()
item = self.get()
callback = None
if not item:
logging.info('NewTimer wait')
self.cond.wait()
elif item[0] <= time.time():
new_item = self.pop()
callback = new_item[1]
else:
logging.info('NewTimer start sys timer and wait')
self.timer = threading.Timer(item[0]-time.time(),self.execute)
self.timer.start()
self.cond.wait()
self.cond.release()
if callback:
callback(item[0])
def add(self, item):
# print('add', item)
self.cond.acquire()
item[0] = item[0] + time.time()
self.queues.append(item)
self.queues.sort(key=lambda x: x[0])
logging.info('NewTimer add notify')
if self.timer:
self.timer.cancel()
self.timer = None
self.cond.notify()
self.cond.release()
def pop(self):
item = None
if len(self.queues) > 0:
item = self.queues.pop(0)
return item
def get(self):
item = None
if len(self.queues) > 0:
item = self.queues[0]
return item
def execute(self):
logging.info('NewTimer execute notify')
self.cond.acquire()
self.cond.notify()
self.cond.release()
执行及输出:
if __name__ == '__main__': def func(): while True: print(threading.active_count()) time.sleep(1) f1 = threading.Thread(target=func) f1.start() logging.basicConfig(level=logging.INFO,format="%(asctime)s %(message)s", datefmt="%m/%d/%Y %H:%M:%S [%A]") newtimer = NewTimer() newtimer.start() def func1(*args): logging.info('func1 %s'%args) time.sleep(5) def func2(*args): logging.info('func2 %s' % args) time.sleep(5) def func3(*args): logging.info('func3 %s' % args) time.sleep(5) def func4(*args): logging.info('func4 %s' % args) time.sleep(5) def func5(*args): logging.info('func5 %s' % args) time.sleep(5) newtimer.add([5,func1]) newtimer.add([4,func2]) newtimer.add([3,func3]) newtimer.add([2,func4]) newtimer.add([1,func5]) time.sleep(1) newtimer.add([1,func1]) newtimer.add([2,func2]) newtimer.add([3,func3]) newtimer.add([4,func4]) newtimer.add([5,func5]) # 输出 # 2 # 07/27/2017 11:26:19 [Thursday] NewTimer wait # 07/27/2017 11:26:19 [Thursday] NewTimer add notify # 07/27/2017 11:26:19 [Thursday] NewTimer add notify # 07/27/2017 11:26:19 [Thursday] NewTimer add notify # 07/27/2017 11:26:19 [Thursday] NewTimer add notify # 07/27/2017 11:26:19 [Thursday] NewTimer add notify # 07/27/2017 11:26:19 [Thursday] NewTimer start sys timer and wait # 07/27/2017 11:26:20 [Thursday] NewTimer execute notify # 4 # 07/27/2017 11:26:20 [Thursday] func5 1501125980.2175007 # 07/27/2017 11:26:20 [Thursday] NewTimer add notify # 07/27/2017 11:26:20 [Thursday] NewTimer add notify # 07/27/2017 11:26:20 [Thursday] NewTimer add notify # 07/27/2017 11:26:20 [Thursday] NewTimer add notify # 07/27/2017 11:26:20 [Thursday] NewTimer add notify # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:25 [Thursday] func4 1501125981.2175007 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:30 [Thursday] func1 1501125981.218279 # 3 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:35 [Thursday] func3 1501125982.2175007 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:40 [Thursday] func2 1501125982.218279 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:45 [Thursday] func2 1501125983.2175007 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:50 [Thursday] func3 1501125983.218279 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:26:55 [Thursday] func1 1501125984.2175007 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:27:00 [Thursday] func4 1501125984.218279 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:27:05 [Thursday] func5 1501125985.218279 # 3 # 3 # 3 # 3 # 3 # 07/27/2017 11:27:10 [Thursday] NewTimer wait
注:这次无论如何测试线程数也不会蹭蹭的上涨,同时可以实现多定时器任务要求;缺点:用到了两线程,没有用到单线程去实现,第二时间精准度问题,需要等待上个定时程序执行完毕,程序才能继续运行