Python Can'；使用mpirun和多处理捕获ctrl-c

好消息：用python3 test.py运行这个测试，然后按ctrl-c。它应该停下来

坏消息：使用

mpirun-n1 python3 test.py运行此测试

按ctrl-c。Ops，mpirun被终止，但multiprocessing.pool生成的所有python进程将永远存在。如何解决这个问题

test.py：

from mpi4py import MPI
import multiprocessing as mp
import signal
import time

class GracefulKiller:
    kill_now = False

    def __init__(self):
        signal.signal(signal.SIGINT, self.exit_gracefully)
        signal.signal(signal.SIGTERM, self.exit_gracefully)
    def exit_gracefully(self, signum, frame):
        self.kill_now = True
        print("I kill")

def worker(e):
    killer = GracefulKiller()
    while(True):
        if killer.kill_now:
            e.set()
        if e.is_set():
            return

def main():
    comm = MPI.COMM_WORLD
    rank = comm.Get_rank()
    size = comm.Get_size()

    killer = GracefulKiller()
    with mp.Manager() as manager:
        e = manager.Event()
        pool = mp.Pool()
        arg = []
        for i in range(100):
            arg += [e]
        r = pool.map_async(worker, arg)
        r.get()
        pool.join()
        pool.close()
        if killer.kill_now:
            e.set()
        if e.is_set():
            comm.Abort()

main()

from mpi4py import MPI
import multiprocessing as mp
import signal
import time
class GracefulKiller:
    kill_now = False

    def __init__(self, name, pool=None):
        signal.signal(signal.SIGINT, self.exit_gracefully)
        signal.signal(signal.SIGTERM, self.exit_gracefully)
        self.name = name
        self.pool = pool
    def exit_gracefully(self, signum, frame):
        self.kill_now = True
        print("I kill.", self.name)
        if self.pool is not None:
            self.pool.close()
            self.pool.terminate()

def worker(e):
    killer = GracefulKiller('worker')
    while(True):
        if killer.kill_now:
            e.set()
        if e.is_set():
            return

def main():
    comm = MPI.COMM_WORLD
    rank = comm.Get_rank()
    size = comm.Get_size()

    with mp.Manager() as manager:
        e = manager.Event()
        pool = mp.Pool()

        killer = GracefulKiller('main', pool)
        arg = []
        for i in range(100):
            arg += [e]
        r = pool.map_async(worker, arg)
        r.get()
        pool.join()
        pool.close()
        if killer.kill_now:
            e.set()
        if e.is_set():
            comm.Abort()

main()

GracefulKiller

来自

mpirun

来自openmpi。我在Ubuntu和CentOS上测试了这个

---

更新：

• 我添加了一行

  print（“I kill”）

  。然后我再次尝试使用

  mpirun

  。它只打印了一次

  I kill

  ，但仍有一堆

  python3

  进程在运行

---

更新2：

• 为优雅杀手添加了一个名字
• 添加了

  pool.terminate（）

  ，当主进程捕获ctrl-c时，尝试杀死主进程的

  pool

  生成的所有进程

test.py：

from mpi4py import MPI
import multiprocessing as mp
import signal
import time

class GracefulKiller:
    kill_now = False

    def __init__(self):
        signal.signal(signal.SIGINT, self.exit_gracefully)
        signal.signal(signal.SIGTERM, self.exit_gracefully)
    def exit_gracefully(self, signum, frame):
        self.kill_now = True
        print("I kill")

def worker(e):
    killer = GracefulKiller()
    while(True):
        if killer.kill_now:
            e.set()
        if e.is_set():
            return

def main():
    comm = MPI.COMM_WORLD
    rank = comm.Get_rank()
    size = comm.Get_size()

    killer = GracefulKiller()
    with mp.Manager() as manager:
        e = manager.Event()
        pool = mp.Pool()
        arg = []
        for i in range(100):
            arg += [e]
        r = pool.map_async(worker, arg)
        r.get()
        pool.join()
        pool.close()
        if killer.kill_now:
            e.set()
        if e.is_set():
            comm.Abort()

main()

from mpi4py import MPI
import multiprocessing as mp
import signal
import time
class GracefulKiller:
    kill_now = False

    def __init__(self, name, pool=None):
        signal.signal(signal.SIGINT, self.exit_gracefully)
        signal.signal(signal.SIGTERM, self.exit_gracefully)
        self.name = name
        self.pool = pool
    def exit_gracefully(self, signum, frame):
        self.kill_now = True
        print("I kill.", self.name)
        if self.pool is not None:
            self.pool.close()
            self.pool.terminate()

def worker(e):
    killer = GracefulKiller('worker')
    while(True):
        if killer.kill_now:
            e.set()
        if e.is_set():
            return

def main():
    comm = MPI.COMM_WORLD
    rank = comm.Get_rank()
    size = comm.Get_size()

    with mp.Manager() as manager:
        e = manager.Event()
        pool = mp.Pool()

        killer = GracefulKiller('main', pool)
        arg = []
        for i in range(100):
            arg += [e]
        r = pool.map_async(worker, arg)
        r.get()
        pool.join()
        pool.close()
        if killer.kill_now:
            e.set()
        if e.is_set():
            comm.Abort()

main()

• python3 test.py（然后是ctrl-c）：

一堆python3进程仍然存在

I kill. worker/main ...
...
  File "test.py", line 20, in exit_gracefully
    self.pool.terminate()
  File "/usr/lib/python3.5/multiprocessing/pool.py", line 505, in terminate
    self._terminate()
  File "/usr/lib/python3.5/multiprocessing/util.py", line 186, in __call__
    res = self._callback(*self._args, **self._kwargs)
  File "/usr/lib/python3.5/multiprocessing/pool.py", line 535, in _terminate_pool
...
I kill. worker/main ...

• mpirun-n 1 python3 test.py（然后ctrl-c）：

  不打印任何内容

  ，一堆python3进程仍在运行

---

可能有帮助的事情：

---

有什么建议吗？主进程接收到信号，但孩子们没有；在MPI下，它们可以在不同的主机上运行。我怀疑

comm.Abort

不是发送终止信号的正确方式，或者

mpirun

使用不同的信号终止子进程

SIGINT

和

SIGTERM

在这方面有所不同，但是Ctrl+C发送SIGINT。你是对的。只有主进程的优雅杀手才能捕捉到ctrl-c。子进程没有做任何事情。仍然找不到修复方法。@你有没有发现这个问题？我也有同样的问题，这很令人沮丧。