多线程脚本最后挂起

时间:2014-08-19 02:31:20

标签: python multithreading queue

您好:我正在尝试让此脚本正常运行。有时取决于用户数量(下面的示例显示3但可以轻松增加),脚本不会退出。所有作业都已完成,但脚本只是挂起而没有退出。我认为我对while True worker中存在的方法是问题,但我不知道另一种选择。有什么想法吗?

import datetime, logging, os.path, queue, random, threading, time

script = os.path.basename(__file__)
logging.basicConfig(level=logging.DEBUG, format="%(asctime)-4s %(thread)6s  %(message)s", datefmt="%m-%d %H:%M:%S",
    filename="%s_%s.log"%(script[:script.find(".")],datetime.datetime.today().strftime("%Y%m%d-%H%M%S")))

class User(object):

    def __init__(self, id, ndelay, mind, maxd):
        self.id = id
        self.numdelay = ndelay #number of delays
        self.mind = mind       #min delay
        self.maxd = maxd       #max delay
        self.currdelaynum = 0  #index for next delay

    def hasDelay(self):
        if self.currdelaynum >= 0 and self.currdelaynum < self.numdelay:
            return True

    def runNextDelay(self):
        delay = round(self.mind + random.random()*(self.maxd - self.mind))
        logging.info("%s beg (delay=%d)"%(self.id,delay))
        time.sleep(delay)
        logging.info("%s end"%self.id)            
        self.currdelaynum += 1


def worker(unext,udone):
    while True:
        if unext.qsize() > 0:
            m = unext.get()
            users_all[m].runNextDelay()
            if users_all[m].hasDelay():
                unext.put(m)
            else:
                udone.put(m)
        else:
            if udone.qsize() >= len(users_all):
                break


if __name__=='__main__':
    random.seed(10)

    #global users_all
    users_all = list()
    users_all.append(User("aa",2,3,9))
    users_all.append(User("bb",3,2,4))
    users_all.append(User("cc",1,4,5))

    users_next = queue.Queue()
    users_done = queue.Queue()
    for n in range(len(users_all)):
        users_next.put(n)

    threads = [threading.Thread(target=worker, args=(users_next,users_done)) for n in range(2)]
    for t in threads: t.start()
    for t in threads: t.join()

大多数多线程python示例都有一个预先知道的作业队列。我正在编写一个脚本来测试在数据库上并行运行的查询的响应时间。为了使上面的示例自包含,我用sleep替换了它的ODBC查询部分。我还要感谢任何有关更好实施的意见。

根据评论更新版本

def worker(unext):
    while True:
        try:
            m = unext.get_nowait()
            users_all[m].runNextDelay()
            if users_all[m].hasDelay():
                unext.put(m)
        except queue.Empty:
            break

3 个答案:

答案 0 :(得分:0)

正如univerio评论的那样,有竞争条件。通常,当处理多个线程之间共享的对象时,请问自己这个问题,如果我的线程在此时被中断并且另一个线程被允许运行会发生什么? univerio概述的情况是qsize()调用可能在线程A中返回非零,然后线程B运行并从同一队列中拉出一个项目。当线程A再次运行以执行get()假设队列中的项目有误时,get()可能会阻塞。

以下是一些未经测试的代码,可用于指导您的最终实施:

def worker(unext, udone):
    while True:
        try:
            m = unext.get_nowait()
            users_all[m].runNextDelay()
            if users_all[m].hasDelay():
                unext.put(m)
            else:
                udone.put(m)
        except queue.Queue.Empty:
            if udone.qsize() >= len(users_all):
                break

这仍然不是一个理想的实现,因为当unext队列为空但其他线程尚未完成处理时,while循环将在所有线程中疯狂旋转,等待最后一个线程完成。 / p>

最好让线程完成工作并在没有剩余工作时退出,让主线程等待udone.qsize() >= len(users_all)条件变为真。

答案 1 :(得分:0)

这是多线程代码的另一个版本。变化:

1)线程有正确的名称(&#34; thread-1&#34;),它们包含在日志中

2)队列保存用户实例,而不是索引到全局数组

3)如果队列中的None为空,则线程会自行停止。初始代码put()将多个用户放入输入队列,然后在每个线程的末尾添加一个无,这样每个线程都会发出信号退出。

4)工人在开始和停止时记录;用户对象可直接打印。

import datetime, logging, os.path, random, sys, threading, time
import Queue as queue

script = os.path.basename(__file__)
logging.basicConfig(
    level=logging.DEBUG, 
    format="%(asctime)-4s %(threadName)s  %(message)s", datefmt="%m-%d %H:%M:%S",
    stream=sys.stderr,
    # filename="%s_%s.log"%(script[:script.find(".")],datetime.datetime.today().strftime("%Y%m%d-%H%M%S")))
)

class User(object):

    def __init__(self, id, ndelay, mind, maxd):
        self.id = id
        self.numdelay = ndelay #number of delays
        self.mind = mind       #min delay
        self.maxd = maxd       #max delay
        self.currdelaynum = 0  #index for next delay

    def __repr__(self):
        return '<User: id={}>'.format(self.id)

    def hasDelay(self):
        return (
            self.currdelaynum >= 0 
            and self.currdelaynum < self.numdelay
        )

    def runNextDelay(self):
        delay = round(self.mind + random.random()*(self.maxd - self.mind))
        logging.info("%s beg (delay=%d)", self.id, delay)
        time.sleep(delay)
        logging.info("%s end", self.id)
        self.currdelaynum += 1


def worker(unext, udone):
    logging.info('start')
    for user in iter(unext.get, None):
        while True:
            user.runNextDelay()
            if not user.hasDelay():
                break
            logging.debug('%s: reloop', user)
        udone.put(user)
    logging.info('done')


if __name__=='__main__':
    random.seed(10)

    users_all = list()
    users_all.append(User("aa",2,3,9))
    users_all.append(User("bb",3,2,4))
    users_all.append(User("cc",1,4,5))

    users_next = queue.Queue()
    users_done = queue.Queue()
    for user in users_all:
        users_next.put(user)

    # flag each thread to exit at end
    num_threads = 2
    for _ in range(num_threads):
        users_next.put(None)

    threads = [
        threading.Thread(
            target=worker, 
            args=(users_next,users_done),
            name='thread-{}'.format(n),
        )
        for n in range(num_threads)
    ]
    for t in threads:
        t.start()
    for t in threads:
        t.join()

输出

08-19 12:29:29 thread-0  start
08-19 12:29:29 thread-0  aa beg (delay=6)
08-19 12:29:29 thread-1  start
08-19 12:29:29 thread-1  bb beg (delay=3)
08-19 12:29:32 thread-1  bb end
08-19 12:29:32 thread-1  <User: id=bb>: reloop
08-19 12:29:32 thread-1  bb beg (delay=3)
08-19 12:29:35 thread-0  aa end
08-19 12:29:35 thread-0  <User: id=aa>: reloop
08-19 12:29:35 thread-0  aa beg (delay=4)
08-19 12:29:35 thread-1  bb end
08-19 12:29:35 thread-1  <User: id=bb>: reloop
08-19 12:29:35 thread-1  bb beg (delay=4)
08-19 12:29:39 thread-1  bb end
08-19 12:29:39 thread-0  aa end
08-19 12:29:39 thread-0  cc beg (delay=5)
08-19 12:29:39 thread-1  done
08-19 12:29:44 thread-0  cc end
08-19 12:29:44 thread-0  done

答案 2 :(得分:0)

我的工作队列遇到了类似的问题。我的解决方案是(上面提到的)调用get()函数的超时时间为0:

def run(self):
    while not self._stopevent.isSet():
        try:
            self._execute_job_function()
        except queue.Empty:
            pass    #make sure the application doesn't crash when the jobqueue is empty

def _execute_job_function(self):
    job = self._job_list.get(False, 0)  #calling get function with time-out = 0 to prevent hanging

    print("Executing job: {0}".format(job))
    self._results_queue.put("{0} - Done".format(job))
    self._job_list.task_done()

我希望这会对你有所帮助。