timeout on subprocess readline in python

问题:

I have a small issue that I'm not quite sure how to solve. Here is a minimal example:

What I have

scan_process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
while(some_criterium):
    line = scan_process.stdout.readline()
    some_criterium = do_something(line)

What I would like

scan_process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
while(some_criterium):
    line = scan_process.stdout.readline()
    if nothing_happens_after_10s:
        break
    else:
        some_criterium = do_something(line)

I read a line from a subprocess and do something with it. What I want is to exit if no line arrived after a fixed time interval. Any recommendations?

回答1:

Thanks for all the answers! I found a way to solve my problem by simply using select.poll to peek into stdout.

import select
...
scan_process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
poll_obj = select.poll()
poll_obj.register(scan_process.stdout, select.POLLIN)   
while(some_criterium and not time_limit):
    poll_result = poll_obj.poll(0)
    if poll_result:
        line = scan_process.stdout.readline()
        some_criterium = do_something(line)
    update(time_limit)

回答2:

Here's a portable solution that enforces the timeout for reading a single line using asyncio:

#!/usr/bin/env python3
import asyncio
import sys
from asyncio.subprocess import PIPE, STDOUT

async def run_command(*args, timeout=None):
    # start child process
    # NOTE: universal_newlines parameter is not supported
    process = await asyncio.create_subprocess_exec(*args,
            stdout=PIPE, stderr=STDOUT)

    # read line (sequence of bytes ending with b'\n') asynchronously
    while True:
        try:
            line = await asyncio.wait_for(process.stdout.readline(), timeout)
        except asyncio.TimeoutError:
            pass
        else:
            if not line: # EOF
                break
            elif do_something(line): 
                continue # while some criterium is satisfied
        process.kill() # timeout or some criterium is not satisfied
        break
    return await process.wait() # wait for the child process to exit


if sys.platform == "win32":
    loop = asyncio.ProactorEventLoop() # for subprocess' pipes on Windows
    asyncio.set_event_loop(loop)
else:
    loop = asyncio.get_event_loop()

returncode = loop.run_until_complete(run_command("cmd", "arg 1", "arg 2",
                                                 timeout=10))
loop.close()

回答3:

I used something a bit more general in python (IIRC also pieced together from SO questions, but I cannot recall which ones).

import thread
from threading import Timer

def run_with_timeout(timeout, default, f, *args, **kwargs):
    if not timeout:
        return f(*args, **kwargs)
    try:
        timeout_timer = Timer(timeout, thread.interrupt_main)
        timeout_timer.start()
        result = f(*args, **kwargs)
        return result
    except KeyboardInterrupt:
        return default
    finally:
        timeout_timer.cancel()

Be warned, though, this uses an interrupt to stop whatever function you give it. This might not be a good idea for all functions and it also prevents you from closing the program with ctrl+c during the timeout (i.e. ctrl+c will be handled as a timeout) You could use this an call it like:

scan_process = subprocess.Popen(command, stdout=subprocess.PIPE, stderr=subprocess.STDOUT)
while(some_criterium):
    line = run_with_timeout(timeout, None, scan_process.stdout.readline)
    if line is None:
        break
    else:
        some_criterium = do_something(line)

Might be a bit overkill, though. I suspect there is a simpler option for your case that I don't know.

回答4:

In Python 3, a timeout option has been added to the subprocess module. Using a structure like

try:
    o, e = process.communicate(timeout=10)
except TimeoutExpired:
    process.kill()
    o, e = process.communicate()

analyze(o)

would be a proper solution.

Since the output is expected to contain a new line character, it's safe to assume that it is text (as in printable, readable), in which case universal_newlines=True flag is strongly recommended.

If Python2 is a must, please use https://pypi.python.org/pypi/subprocess32/ (backport)

For a pure-python Python 2 solution, look at Using module 'subprocess' with timeout.

回答5:

Try using signal.alarm:

#timeout.py
import signal,sys

def timeout(sig,frm):
  print "This is taking too long..."
  sys.exit(1)

signal.signal(signal.SIGALRM, timeout)
signal.alarm(10)
byte=0

while 'IT' not in open('/dev/urandom').read(2):
  byte+=2
print "I got IT in %s byte(s)!" % byte

A couple of runs to show it works:

$ python timeout.py 
This is taking too long...
$ python timeout.py 
I got IT in 4672 byte(s)!

For more detailed example see pGuides.

回答6:

A portable solution is to use a thread to kill the child process if reading a line takes too long:

#!/usr/bin/env python3
from subprocess import Popen, PIPE, STDOUT

timeout = 10
with Popen(command, stdout=PIPE, stderr=STDOUT,
           universal_newlines=True) as process:  # text mode
    # kill process in timeout seconds unless the timer is restarted
    watchdog = WatchdogTimer(timeout, callback=process.kill, daemon=True)
    watchdog.start()
    for line in process.stdout:
        # don't invoke the watcthdog callback if do_something() takes too long
        with watchdog.blocked:
            if not do_something(line):  # some criterium is not satisfied
                process.kill()
                break
            watchdog.restart()  # restart timer just before reading the next line
    watchdog.cancel()

where WatchdogTimer class is like threading.Timer that can be restarted and/or blocked:

from threading import Event, Lock, Thread
from subprocess import Popen, PIPE, STDOUT
from time import monotonic  # use time.time or monotonic.monotonic on Python 2

class WatchdogTimer(Thread):
    """Run *callback* in *timeout* seconds unless the timer is restarted."""

    def __init__(self, timeout, callback, *args, timer=monotonic, **kwargs):
        super().__init__(**kwargs)
        self.timeout = timeout
        self.callback = callback
        self.args = args
        self.timer = timer
        self.cancelled = Event()
        self.blocked = Lock()

    def run(self):
        self.restart() # don't start timer until `.start()` is called
        # wait until timeout happens or the timer is canceled
        while not self.cancelled.wait(self.deadline - self.timer()):
            # don't test the timeout while something else holds the lock
            # allow the timer to be restarted while blocked
            with self.blocked:
                if self.deadline <= self.timer() and not self.cancelled.is_set():
                    return self.callback(*self.args)  # on timeout

    def restart(self):
        """Restart the watchdog timer."""
        self.deadline = self.timer() + self.timeout

    def cancel(self):
        self.cancelled.set()

回答7:

while your (Tom's) solution works, using select() in the C idiom is more compact. this is the equivalent of your answer

from select import select
scan_process = subprocess.Popen(command, 
                                stdout=subprocess.PIPE,
                                stderr=subprocess.STDOUT,
                                bufsize=1)  # line buffered
while some_criterium and not time_limit:
    poll_result = select([scan_process.stdout], [], [], time_limit)[0]

the rest is the same.

see pydoc select.select.

[Note: this is Unix-specific, as are some of the other answers.]

[Note 2: edited to add line buffering as per OP request]

[Note 3: the line buffering may not be reliable in all circumstances, leading to readline() blocking]