Python多处理子项在运行退出

时间:2015-08-12 12:55:23

标签: python multiprocessing

我正在使用多处理来调试问题 我有以下孩子:

class Target(multiprocessing.Process):
    def __init__(self, work_queue, result_queue, suite_name, test_settings, html_log_dir, output_file, ip_address, fit_dir):
        multiprocessing.Process.__init__(self)

        # initialize other variables

    def run(self):
        print multiprocessing.current_process()
        suite_start_time = time.clock()
        while not self.kill_received:
            # get a task
            try:
                job = self.work_queue.get(True, 2)
            except Queue.Empty:
                self._log('Work queue empty, creating XML result file')
                self.create_xml_result_file(suite_start_time)
                break

            # the actual processing, run the test.

        fitnesse_common.log_output("\n(PID " + str(self.pid) + "): End of process")

    def create_xml_result_file(self, start_time):
        # generate result 

父进程基本上只启动了几个(12)目标并等待它们全部加入。

问题是由于某种原因,子进程正在运行到run函数的末尾(我看到进程打印结束),然后由于某种原因没有终止,这阻止了父进程的继续。

编辑 - 并非所有衍生进程都挂起,只有几个。在12个衍生过程中,通常只有2-4个过程在完成其运行功能后挂起。

我考虑在run函数结束时调用terminate,但Python文档表明这是一个坏主意。

我已经看过有关Python多处理的Stack Overflow的几篇不同文章,其中大部分与父进程的问题有关。

非常感谢任何想法或帮助。

更新:这是一个很容易复制问题的脚本:

import multiprocessing, Queue
import subprocess
import time
import sys

class Target(multiprocessing.Process):
    def __init__(self, work_queue, results_queue, delay_length):
        # base class initialization
        multiprocessing.Process.__init__(self)

        # job management stuff
        self.work_queue = work_queue
        self.results_queue = results_queue
        self.delay_length = delay_length
        self.kill_received = False

    def run(self):
        while not self.kill_received:
            # get a task
            try:
                job = self.work_queue.get(True, 2)
            except Queue.Empty:
                self._log('Work queue empty, prepare to terminate')
                break

            time.sleep(self.delay_length)
            self._log("Sleeping done")

            results = self._run_an_application(job)

            self.results_queue.put(results)
            self._log("Have put results on queue " + str(job) + "-" + results)
        self._log("\n(PID " + str(self.pid) + "): End of process")
    def _log(self, text):
        print ('PID ' + str(self.pid) + ' => ' + text)
        sys.stdout.flush()

    def _run_an_application(self, app):
        try: 
            test_output = subprocess.check_output(app)
        except subprocess.CalledProcessError, e:
            log_output('### Process check_output threw exception CalledProcessError')
            test_output = e.output

        return test_output

if __name__ == "__main__":

    test_jobs = []
    started_targets = []

    # run
    # load up work queue
    for i in range(500):
        test_jobs.append('spewage')
    work_queue = multiprocessing.Queue()

    for job in test_jobs:
        work_queue.put(job)

    # create a queue to pass to targets to store the results
    result_queue = multiprocessing.Queue()

    # spawn targets
    for i in range(12):
        started_targets.append(Target(work_queue, result_queue, i))

    # start all targets
    for i in range(len(started_targets)):
        started_targets[i].start()
        print "starting process no %s with id: %s" % (i, started_targets[i].pid)

    print 'Waiting for all processes to join'

    # wait for all targets to finish
    for i in range(len(started_targets)):
        started_targets[i].join()

    print 'All processes have joined'

    # collect the results off the queue
    while not result_queue.empty():
        target_result = result_queue.get()
        print "Test job - " + target_result
    print ('All Tests completed')

这是“spewage”应用程序的源代码(它是C ++)。

#include <iostream>
#include <windows.h>
using namespace std;

int main()
{
    for (int i = 0; i < 500; i++)
    {
        cout << "abcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" << endl;
        Sleep(20); 
    }
    return 0;
}

由于它似乎与推送到stdout的数量有关,因此C ++程序很容易被另一个打印很多东西的脚本所取代。

1 个答案:

答案 0 :(得分:1)

我设法解决了这个问题。它似乎与子流程中的输出量有关。在进程'run()函数结束时,我需要放置self.results_queue.cancel_join_thread()

我仍然很好奇为什么它在没有很多stdout的情况下工作,但是当有时,进程会挂起。根据Python文档,我使用result_queue的方式应该一直锁定,即使它没有。