将模块“子进程”与超时一起使用

Here’s the Python code to run an arbitrary command returning its stdout data, or raise an exception on non-zero exit codes:

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

communicate is used to wait for the process to exit:

stdoutdata, stderrdata = proc.communicate()

The subprocess module does not support timeout–ability to kill a process running for more than X number of seconds–therefore, communicate may take forever to run.

What is the simplest way to implement timeouts in a Python program meant to run on Windows and Linux?

In Python 3.3+:

from subprocess import STDOUT, check_output

output = check_output(cmd, stderr=STDOUT, timeout=seconds)

output is a byte string that contains command’s merged stdout, stderr data.

raises CalledProcessError on non-zero exit status as specified in the question’s text unlike proc.communicate() method.

I’ve removed shell=True because it is often used unnecessarily. You can always add it back if cmd indeed requires it. If you add shell=True i.e., if the child process spawns its own descendants; check_output() can return much later than the timeout indicates, see Subprocess timeout failure.

The timeout feature is available on Python 2.x via the subprocess32 backport of the 3.2+ subprocess module.

I don’t know much about the low level details; but, given that in python 2.6 the API offers the ability to wait for threads and terminate processes, what about running the process in a separate thread?

import subprocess, threading

class Command(object):
    def __init__(self, cmd):
        self.cmd = cmd
        self.process = None

    def run(self, timeout):
        def target():
            print 'Thread started'
            self.process = subprocess.Popen(self.cmd, shell=True)
            self.process.communicate()
            print 'Thread finished'

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

        thread.join(timeout)
        if thread.is_alive():
            print 'Terminating process'
            self.process.terminate()
            thread.join()
        print self.process.returncode

command = Command("echo 'Process started'; sleep 2; echo 'Process finished'")
command.run(timeout=3)
command.run(timeout=1)

The output of this snippet in my machine is:

Thread started
Process started
Process finished
Thread finished
0
Thread started
Process started
Terminating process
Thread finished
-15

where it can be seen that, in the first execution, the process finished correctly (return code 0), while the in the second one the process was terminated (return code -15).

I haven’t tested in windows; but, aside from updating the example command, I think it should work since I haven’t found in the documentation anything that says that thread.join or process.terminate is not supported.

jcollado’s answer can be simplified using the threading.Timer class:

import shlex
from subprocess import Popen, PIPE
from threading import Timer

def run(cmd, timeout_sec):
    proc = Popen(shlex.split(cmd), stdout=PIPE, stderr=PIPE)
    timer = Timer(timeout_sec, proc.kill)
    try:
        timer.start()
        stdout, stderr = proc.communicate()
    finally:
        timer.cancel()

# Examples: both take 1 second
run("sleep 1", 5)  # process ends normally at 1 second
run("sleep 5", 1)  # timeout happens at 1 second

If you’re on Unix,

import signal
  ...
class Alarm(Exception):
    pass

def alarm_handler(signum, frame):
    raise Alarm

signal.signal(signal.SIGALRM, alarm_handler)
signal.alarm(5*60)  # 5 minutes
try:
    stdoutdata, stderrdata = proc.communicate()
    signal.alarm(0)  # reset the alarm
except Alarm:
    print "Oops, taking too long!"
    # whatever else

Here is Alex Martelli’s solution as a module with proper process killing. The other approaches do not work because they do not use proc.communicate(). So if you have a process that produces lots of output, it will fill its output buffer and then block until you read something from it.

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)

I’ve modified sussudio answer. Now function returns: (returncode, stdout, stderr, timeout) – stdout and stderr is decoded to utf-8 string

def kill_proc(proc, timeout):
  timeout["value"] = True
  proc.kill()

def run(cmd, timeout_sec):
  proc = subprocess.Popen(shlex.split(cmd), stdout=subprocess.PIPE, stderr=subprocess.PIPE)
  timeout = {"value": False}
  timer = Timer(timeout_sec, kill_proc, [proc, timeout])
  timer.start()
  stdout, stderr = proc.communicate()
  timer.cancel()
  return proc.returncode, stdout.decode("utf-8"), stderr.decode("utf-8"), timeout["value"]

surprised nobody mentioned using timeout

timeout 5 ping -c 3 somehost

This won’t for work for every use case obviously, but if your dealing with a simple script, this is hard to beat.

Also available as gtimeout in coreutils via homebrew for mac users.

by call() and communicate() in the subprocess module (as of Python3.3):

import subprocess

subprocess.call("command", timeout=20, shell=True)

This will call the command and raise the exception

subprocess.TimeoutExpired

if the command doesn’t finish after 20 seconds.

You can then handle the exception to continue your code, something like:

try:
    subprocess.call("command", timeout=20, shell=True)
except subprocess.TimeoutExpired:
    # insert code here

Hope this helps.

Another option is to write to a temporary file to prevent the stdout blocking instead of needing to poll with communicate(). This worked for me where the other answers did not; for example on windows.

    outFile =  tempfile.SpooledTemporaryFile() 
    errFile =   tempfile.SpooledTemporaryFile() 
    proc = subprocess.Popen(args, stderr=errFile, stdout=outFile, universal_newlines=False)
    wait_remaining_sec = timeout

    while proc.poll() is None and wait_remaining_sec > 0:
        time.sleep(1)
        wait_remaining_sec -= 1

    if wait_remaining_sec <= 0:
        killProc(proc.pid)
        raise ProcessIncompleteError(proc, timeout)

    # read temp streams from start
    outFile.seek(0);
    errFile.seek(0);
    out = outFile.read()
    err = errFile.read()
    outFile.close()
    errFile.close()

I don’t know why it isn’t mentionned but since Python 3.5, there’s a new universal command (that is meant to replace check_call, check_output …) and which has the timeout parameter as well.

subprocess.run(args, *, stdin=None, input=None, stdout=None, stderr=None, shell=False, cwd=None, timeout=None, check=False, encoding=None, errors=None)

Run the command described by args. Wait for command to complete, then return a CompletedProcess instance.

It raises a subprocess.TimeoutExpired exception when the timeout is expired.

Here is my solution, I was using Thread and Event:

import subprocess
from threading import Thread, Event

def kill_on_timeout(done, timeout, proc):
    if not done.wait(timeout):
        proc.kill()

def exec_command(command, timeout):

    done = Event()
    proc = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.PIPE)

    watcher = Thread(target=kill_on_timeout, args=(done, timeout, proc))
    watcher.daemon = True
    watcher.start()

    data, stderr = proc.communicate()
    done.set()

    return data, stderr, proc.returncode

In action:

In [2]: exec_command(['sleep', '10'], 5)
Out[2]: ('', '', -9)

In [3]: exec_command(['sleep', '10'], 11)
Out[3]: ('', '', 0)

The solution I use is to prefix the shell command with timelimit. If the comand takes too long, timelimit will stop it and Popen will have a returncode set by timelimit. If it is > 128, it means timelimit killed the process.

See also python subprocess with timeout and large output (>64K)

I added the solution with threading from jcollado to my Python module easyprocess.

Install:

pip install easyprocess

Example:

from easyprocess import Proc

# shell is not supported!
stdout=Proc('ping localhost').call(timeout=1.5).stdout
print stdout

if you are using python 2, give it a try

import subprocess32

try:
    output = subprocess32.check_output(command, shell=True, timeout=3)
except subprocess32.TimeoutExpired as e:
    print e

Prepending the Linux command timeout isn’t a bad workaround and it worked for me.

cmd = "timeout 20 "+ cmd
subprocess.Popen(cmd.split(), stdout=subprocess.PIPE, stderr=subprocess.PIPE)
(output, err) = p.communicate()

I’ve implemented what I could gather from a few of these. This works in Windows, and since this is a community wiki, I figure I would share my code as well:

class Command(threading.Thread):
    def __init__(self, cmd, outFile, errFile, timeout):
        threading.Thread.__init__(self)
        self.cmd = cmd
        self.process = None
        self.outFile = outFile
        self.errFile = errFile
        self.timed_out = False
        self.timeout = timeout

    def run(self):
        self.process = subprocess.Popen(self.cmd, stdout = self.outFile, \
            stderr = self.errFile)

        while (self.process.poll() is None and self.timeout > 0):
            time.sleep(1)
            self.timeout -= 1

        if not self.timeout > 0:
            self.process.terminate()
            self.timed_out = True
        else:
            self.timed_out = False

Then from another class or file:

        outFile =  tempfile.SpooledTemporaryFile()
        errFile =   tempfile.SpooledTemporaryFile()

        executor = command.Command(c, outFile, errFile, timeout)
        executor.daemon = True
        executor.start()

        executor.join()
        if executor.timed_out:
            out = 'timed out'
        else:
            outFile.seek(0)
            errFile.seek(0)
            out = outFile.read()
            err = errFile.read()

        outFile.close()
        errFile.close()

Once you understand full process running machinery in *unix, you will easily find simplier solution:

Consider this simple example how to make timeoutable communicate() meth using select.select() (available alsmost everythere on *nix nowadays). This also can be written with epoll/poll/kqueue, but select.select() variant could be a good example for you. And major limitations of select.select() (speed and 1024 max fds) are not applicapable for your task.

This works under *nix, does not create threads, does not uses signals, can be lauched from any thread (not only main), and fast enought to read 250mb/s of data from stdout on my machine (i5 2.3ghz).

There is a problem in join’ing stdout/stderr at the end of communicate. If you have huge program output this could lead to big memory usage. But you can call communicate() several times with smaller timeouts.

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)

You can do this using select

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

I’ve used killableprocess successfully on Windows, Linux and Mac. If you are using Cygwin Python, you’ll need OSAF’s version of killableprocess because otherwise native Windows processes won’t get killed.

Although I haven’t looked at it extensively, this decorator I found at ActiveState seems to be quite useful for this sort of thing. Along with subprocess.Popen(..., close_fds=True), at least I’m ready for shell-scripting in Python.

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()

There’s an idea to subclass the Popen class and extend it with some simple method decorators. Let’s call it ExpirablePopen.

from logging import error
from subprocess import Popen
from threading import Event
from threading import Thread


class ExpirablePopen(Popen):

    def __init__(self, *args, **kwargs):
        self.timeout = kwargs.pop('timeout', 0)
        self.timer = None
        self.done = Event()

        Popen.__init__(self, *args, **kwargs)

    def __tkill(self):
        timeout = self.timeout
        if not self.done.wait(timeout):
            error('Terminating process {} by timeout of {} secs.'.format(self.pid, timeout))
            self.kill()

    def expirable(func):
        def wrapper(self, *args, **kwargs):
            # zero timeout means call of parent method
            if self.timeout == 0:
                return func(self, *args, **kwargs)

            # if timer is None, need to start it
            if self.timer is None:
                self.timer = thr = Thread(target=self.__tkill)
                thr.daemon = True
                thr.start()

            result = func(self, *args, **kwargs)
            self.done.set()

            return result
        return wrapper

    wait = expirable(Popen.wait)
    communicate = expirable(Popen.communicate)


if __name__ == '__main__':
    from subprocess import PIPE

    print ExpirablePopen('ssh -T git@bitbucket.org', stdout=PIPE, timeout=1).communicate()

I had the problem that I wanted to terminate a multithreading subprocess if it took longer than a given timeout length. I wanted to set a timeout in Popen(), but it did not work. Then, I realized that Popen().wait() is equal to call() and so I had the idea to set a timeout within the .wait(timeout=xxx) method, which finally worked. Thus, I solved it this way:

import os
import sys
import signal
import subprocess
from multiprocessing import Pool

cores_for_parallelization = 4
timeout_time = 15  # seconds

def main():
    jobs = [...YOUR_JOB_LIST...]
    with Pool(cores_for_parallelization) as p:
        p.map(run_parallel_jobs, jobs)

def run_parallel_jobs(args):
    # Define the arguments including the paths
    initial_terminal_command = 'C:\\Python34\\python.exe'  # Python executable
    function_to_start = 'C:\\temp\\xyz.py'  # The multithreading script
    final_list = [initial_terminal_command, function_to_start]
    final_list.extend(args)

    # Start the subprocess and determine the process PID
    subp = subprocess.Popen(final_list)  # starts the process
    pid = subp.pid

    # Wait until the return code returns from the function by considering the timeout. 
    # If not, terminate the process.
    try:
        returncode = subp.wait(timeout=timeout_time)  # should be zero if accomplished
    except subprocess.TimeoutExpired:
        # Distinguish between Linux and Windows and terminate the process if 
        # the timeout has been expired
        if sys.platform == 'linux2':
            os.kill(pid, signal.SIGTERM)
        elif sys.platform == 'win32':
            subp.terminate()

if __name__ == '__main__':
    main()

Unfortunately, I’m bound by very strict policies on the disclosure of source code by my employer, so I can’t provide actual code. But for my taste the best solution is to create a subclass overriding Popen.wait() to poll instead of wait indefinitely, and Popen.__init__ to accept a timeout parameter. Once you do that, all the other Popen methods (which call wait) will work as expected, including communicate.

https://pypi.python.org/pypi/python-subprocess2 provides extensions to the subprocess module which allow you to wait up to a certain period of time, otherwise terminate.

So, to wait up to 10 seconds for the process to terminate, otherwise kill:

pipe  = subprocess.Popen('...')

timeout =  10

results = pipe.waitOrTerminate(timeout)

This is compatible with both windows and unix. “results” is a dictionary, it contains “returnCode” which is the return of the app (or None if it had to be killed), as well as “actionTaken”. which will be “SUBPROCESS2_PROCESS_COMPLETED” if the process completed normally, or a mask of “SUBPROCESS2_PROCESS_TERMINATED” and SUBPROCESS2_PROCESS_KILLED depending on action taken (see documentation for full details)

for python 2.6+, use 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

python 2.7

import time
import subprocess

def run_command(cmd, timeout=0):
    start_time = time.time()
    df = subprocess.Popen(cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
    while timeout and df.poll() == None:
        if time.time()-start_time >= timeout:
            df.kill()
            return -1, ""
    output = '\n'.join(df.communicate()).strip()
    return df.returncode, output

import subprocess, optparse, os, sys, re, datetime, threading, time, glob, shutil, xml.dom.minidom, traceback

class OutputManager:
    def __init__(self, filename, mode, console, logonly):
        self.con = console
        self.logtoconsole = True
        self.logtofile = False

        if filename:
            try:
                self.f = open(filename, mode)
                self.logtofile = True
                if logonly == True:
                    self.logtoconsole = False
            except IOError:
                print (sys.exc_value)
                print ("Switching to console only output...\n")
                self.logtofile = False
                self.logtoconsole = True

    def write(self, data):
        if self.logtoconsole == True:
            self.con.write(data)
        if self.logtofile == True:
            self.f.write(data)
        sys.stdout.flush()

def getTimeString():
        return time.strftime("%Y-%m-%d", time.gmtime())

def runCommand(command):
    '''
    Execute a command in new thread and return the
    stdout and stderr content of it.
    '''
    try:
        Output = subprocess.Popen(command, stdout=subprocess.PIPE, shell=True).communicate()[0]
    except Exception as e:
        print ("runCommand failed :%s" % (command))
        print (str(e))
        sys.stdout.flush()
        return None
    return Output

def GetOs():
    Os = ""
    if sys.platform.startswith('win32'):
        Os = "win"
    elif sys.platform.startswith('linux'):
        Os = "linux"
    elif sys.platform.startswith('darwin'):
        Os = "mac"
    return Os


def check_output(*popenargs, **kwargs):
    try:
        if 'stdout' in kwargs: 
            raise ValueError('stdout argument not allowed, it will be overridden.') 

        # Get start time.
        startTime = datetime.datetime.now()
        timeoutValue=3600

        cmd = popenargs[0]

        if sys.platform.startswith('win32'):
            process = subprocess.Popen( cmd, stdout=subprocess.PIPE, shell=True) 
        elif sys.platform.startswith('linux'):
            process = subprocess.Popen( cmd , stdout=subprocess.PIPE, shell=True ) 
        elif sys.platform.startswith('darwin'):
            process = subprocess.Popen( cmd , stdout=subprocess.PIPE, shell=True ) 

        stdoutdata, stderrdata = process.communicate( timeout = timeoutValue )
        retcode = process.poll()

        ####################################
        # Catch crash error and log it.
        ####################################
        OutputHandle = None
        try:
            if retcode >= 1:
                OutputHandle = OutputManager( 'CrashJob_' + getTimeString() + '.txt', 'a+', sys.stdout, False)
                OutputHandle.write( cmd )
                print (stdoutdata)
                print (stderrdata)
                sys.stdout.flush()
        except Exception as e:
            print (str(e))

    except subprocess.TimeoutExpired:
            ####################################
            # Catch time out error and log it.
            ####################################
            Os = GetOs()
            if Os == 'win':
                killCmd = "taskkill /FI \"IMAGENAME eq {0}\" /T /F"
            elif Os == 'linux':
                killCmd = "pkill {0)"
            elif Os == 'mac':
                # Linux, Mac OS
                killCmd = "killall -KILL {0}"

            runCommand(killCmd.format("java"))
            runCommand(killCmd.format("YouApp"))

            OutputHandle = None
            try:
                OutputHandle = OutputManager( 'KillJob_' + getTimeString() + '.txt', 'a+', sys.stdout, False)
                OutputHandle.write( cmd )
            except Exception as e:
                print (str(e))
    except Exception as e:
            for frame in traceback.extract_tb(sys.exc_info()[2]):
                        fname,lineno,fn,text = frame
                        print "Error in %s on line %d" % (fname, lineno)

Was just trying to write something simpler.

#!/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()