下面是运行任意命令返回其标准输出数据的Python代码,或在非零退出码上引发异常:

proc = subprocess.Popen(
    cmd,
    stderr=subprocess.STDOUT,  # Merge stdout and stderr
    stdout=subprocess.PIPE,
    shell=True)

communication用于等待进程退出:

stdoutdata, stderrdata = proc.communicate()

子进程模块不支持超时——杀死运行超过X秒的进程的能力——因此,通信可能需要很长时间才能运行。

在Windows和Linux上运行的Python程序中实现超时的最简单方法是什么?


当前回答

一旦你理解了在*unix中运行机器的整个过程,你将很容易找到更简单的解决方案:

考虑这个简单的例子,如何使用select.select()(现在几乎在*nix上随处可见)创建可超时的communication()冰毒。这也可以用epoll/poll/kqueue来编写,但select.select()变体可能是一个很好的例子。select.select()的主要限制(速度和1024 max fds)不适用于您的任务。

它在*nix下工作,不创建线程,不使用信号,可以从任何线程启动(不仅仅是主线程),并且足够快,可以从我机器上的stdout读取250mb/s的数据(i5 2.3ghz)。

在通信结束时连接stdout/stderr时出现问题。如果你有大量的程序输出,这可能会导致大量的内存使用。但是您可以多次调用communication(),超时时间较小。

class Popen(subprocess.Popen):
    def communicate(self, input=None, timeout=None):
        if timeout is None:
            return subprocess.Popen.communicate(self, input)

        if self.stdin:
            # Flush stdio buffer, this might block if user
            # has been writing to .stdin in an uncontrolled
            # fashion.
            self.stdin.flush()
            if not input:
                self.stdin.close()

        read_set, write_set = [], []
        stdout = stderr = None

        if self.stdin and input:
            write_set.append(self.stdin)
        if self.stdout:
            read_set.append(self.stdout)
            stdout = []
        if self.stderr:
            read_set.append(self.stderr)
            stderr = []

        input_offset = 0
        deadline = time.time() + timeout

        while read_set or write_set:
            try:
                rlist, wlist, xlist = select.select(read_set, write_set, [], max(0, deadline - time.time()))
            except select.error as ex:
                if ex.args[0] == errno.EINTR:
                    continue
                raise

            if not (rlist or wlist):
                # Just break if timeout
                # Since we do not close stdout/stderr/stdin, we can call
                # communicate() several times reading data by smaller pieces.
                break

            if self.stdin in wlist:
                chunk = input[input_offset:input_offset + subprocess._PIPE_BUF]
                try:
                    bytes_written = os.write(self.stdin.fileno(), chunk)
                except OSError as ex:
                    if ex.errno == errno.EPIPE:
                        self.stdin.close()
                        write_set.remove(self.stdin)
                    else:
                        raise
                else:
                    input_offset += bytes_written
                    if input_offset >= len(input):
                        self.stdin.close()
                        write_set.remove(self.stdin)

            # Read stdout / stderr by 1024 bytes
            for fn, tgt in (
                (self.stdout, stdout),
                (self.stderr, stderr),
            ):
                if fn in rlist:
                    data = os.read(fn.fileno(), 1024)
                    if data == '':
                        fn.close()
                        read_set.remove(fn)
                    tgt.append(data)

        if stdout is not None:
            stdout = ''.join(stdout)
        if stderr is not None:
            stderr = ''.join(stderr)

        return (stdout, stderr)

其他回答

只是想写些简单点的东西。

#!/usr/bin/python

from subprocess import Popen, PIPE
import datetime
import time 

popen = Popen(["/bin/sleep", "10"]);
pid = popen.pid
sttime = time.time();
waittime =  3

print "Start time %s"%(sttime)

while True:
    popen.poll();
    time.sleep(1)
    rcode = popen.returncode
    now = time.time();
    if [ rcode is None ]  and  [ now > (sttime + waittime) ] :
        print "Killing it now"
        popen.kill()

下面是Alex Martelli的解决方案,它是一个具有适当进程终止的模块。其他方法不起作用,因为它们没有使用proc. communication()。所以如果你有一个产生大量输出的进程,它会填满它的输出缓冲区,然后阻塞,直到你从中读取一些东西。

from os import kill
from signal import alarm, signal, SIGALRM, SIGKILL
from subprocess import PIPE, Popen

def run(args, cwd = None, shell = False, kill_tree = True, timeout = -1, env = None):
    '''
    Run a command with a timeout after which it will be forcibly
    killed.
    '''
    class Alarm(Exception):
        pass
    def alarm_handler(signum, frame):
        raise Alarm
    p = Popen(args, shell = shell, cwd = cwd, stdout = PIPE, stderr = PIPE, env = env)
    if timeout != -1:
        signal(SIGALRM, alarm_handler)
        alarm(timeout)
    try:
        stdout, stderr = p.communicate()
        if timeout != -1:
            alarm(0)
    except Alarm:
        pids = [p.pid]
        if kill_tree:
            pids.extend(get_process_children(p.pid))
        for pid in pids:
            # process might have died before getting to this line
            # so wrap to avoid OSError: no such process
            try: 
                kill(pid, SIGKILL)
            except OSError:
                pass
        return -9, '', ''
    return p.returncode, stdout, stderr

def get_process_children(pid):
    p = Popen('ps --no-headers -o pid --ppid %d' % pid, shell = True,
              stdout = PIPE, stderr = PIPE)
    stdout, stderr = p.communicate()
    return [int(p) for p in stdout.split()]

if __name__ == '__main__':
    print run('find /', shell = True, timeout = 3)
    print run('find', shell = True)

您可以使用select来完成此操作

import subprocess
from datetime import datetime
from select import select

def call_with_timeout(cmd, timeout):
    started = datetime.now()
    sp = subprocess.Popen(cmd, stdout=subprocess.PIPE)
    while True:
        p = select([sp.stdout], [], [], timeout)
        if p[0]:
            p[0][0].read()
        ret = sp.poll()
        if ret is not None:
            return ret
        if (datetime.now()-started).total_seconds() > timeout:
            sp.kill()
            return None

This solution kills the process tree in case of shell=True, passes parameters to the process (or not), has a timeout and gets the stdout, stderr and process output of the call back (it uses psutil for the kill_proc_tree). This was based on several solutions posted in SO including jcollado's. Posting in response to comments by Anson and jradice in jcollado's answer. Tested in Windows Srvr 2012 and Ubuntu 14.04. Please note that for Ubuntu you need to change the parent.children(...) call to parent.get_children(...).

def kill_proc_tree(pid, including_parent=True):
  parent = psutil.Process(pid)
  children = parent.children(recursive=True)
  for child in children:
    child.kill()
  psutil.wait_procs(children, timeout=5)
  if including_parent:
    parent.kill()
    parent.wait(5)

def run_with_timeout(cmd, current_dir, cmd_parms, timeout):
  def target():
    process = subprocess.Popen(cmd, cwd=current_dir, shell=True, stdout=subprocess.PIPE, stdin=subprocess.PIPE, stderr=subprocess.PIPE)

    # wait for the process to terminate
    if (cmd_parms == ""):
      out, err = process.communicate()
    else:
      out, err = process.communicate(cmd_parms)
    errcode = process.returncode

  thread = Thread(target=target)
  thread.start()

  thread.join(timeout)
  if thread.is_alive():
    me = os.getpid()
    kill_proc_tree(me, including_parent=False)
    thread.join()

对于python 2.6+,使用gevent

 from gevent.subprocess import Popen, PIPE, STDOUT

 def call_sys(cmd, timeout):
      p= Popen(cmd, shell=True, stdout=PIPE)
      output, _ = p.communicate(timeout=timeout)
      assert p.returncode == 0, p. returncode
      return output

 call_sys('./t.sh', 2)

 # t.sh example
 sleep 5
 echo done
 exit 1