Python子进程超时与耗时的任务

时间:2014-04-03 18:40:42

标签: linux python process timeout

以下代码允许运行具有超时的子进程

p = subprocess.Popen([...])
while timeConsumedSoFar < timeoutLimit
    if proc.poll() is not None:
        doSomething
    else:
        time.sleep(2)
os.kill([...])

它适用于常规的linux命令。但是有些命令是耗时的,例如“碎片”

'shred [OPTIONS] FILE [...] '

粉碎逻辑卷可能需要20多分钟。有没有更好的方法来处理这个问题?

1 个答案:

答案 0 :(得分:0)

看起来子进程处理等待中断,所以我无法使用它。

这是一个(可能是错误的)示例,它是一种更传统的unix方法。 注意这不会产生任何来自stdout的输出!实际上它根本不产生输出。

由于所有评论,它看起来比它的更多。

为什么你可以更好地考虑这个问题是因为我们不会继续轮询这个过程。相反,如果命令完成,我们告诉内核在指定的超时之后返回给我们。在任何一种情况下,响应都不会在事件发生后延迟超过几微秒,加上代码只需要设置第一个初始轮询。

#!/usr/bin/python

def runwait(timeout, *args):
  from os import fork, execvp, waitpid, kill, dup2
  from signal import alarm, signal, SIGTERM, SIGALRM, SIG_DFL
  from errno import EINTR

  ## We dont do much in the alarm handler, we only care that we interrupted
  ## the wait() on the process.
  def onalarm(sig, frame):
    pass

  ## This is used for redirecting stdin/out/err to /dev/null
  devnull = open('/dev/null', 'r+')
  pid = fork()
  if not pid:
    ## The child process. Begin by redirecting all input and output.
    dup2(devnull.fileno(), 0)
    dup2(devnull.fileno(), 1)
    dup2(devnull.fileno(), 2)
    ## Probably not a great idea to pass this file descriptor to whatever we 
    ## end up executing.
    devnull.close()

    ## Overwrite child process with new execution context.
    execvp(args[0], args)
    ## If something failed (like command not found) exit 63. WARNING, because we
    ## redirected all output to /dev/null, you WILL NOT be informed the command was
    ## not found directly. Use the exit code to work that out.
    sys.exit(63)

  ## This is the parent process that initiated the fork.
  ## Arm a timer using timeout given by first parameter of function. This
  ## must be an int, not a float. I dont bother checking though cause I'm lazy.    
  signal(SIGALRM, onalarm)
  alarm(timeout)

  ## We wait on the pid, this call typically blocks forever.
  try:
    pid, rc = waitpid(pid, 0)
  except OSError as e:
    ## We will land here if the alarm triggered BEFORE the process completed!
    ## In this case, if we were interrupted to deal with the 
    ## signal handler its definitely an alarm. Otherwise
    ## a peripheral exception occurred (permissions for example) so just re-raise the exception.
    if e.errno == EINTR:
      ## We send a TERM signal to terminate the process and re-wait. This causes
      ## wait to (under normal conditions) come back immediately with the signal 
      ## we just killed it with which parse out further down.
      kill(pid, SIGTERM)
      pid, rc = waitpid(pid, 0)
    else:
      raise

  ## Waits status is a 16bit integer packing a 8bit signal and 8bit return code.
  ## Do some funky bitwise operations to separate the two..
  sig = rc & 0xff
  rc >>= 8
  ## Whatever happened above, always disable the alarm and signal handling.
  alarm(0)
  signal(SIGALRM, SIG_DFL)
  return sig, rc


if __name__ == "__main__":
  # An example when you time out
  print runwait(2, "sleep", "20")
  # An example on success
  print runwait(5, "sleep", "3")
  # More success, but demonstrating no output
  print runwait(5, "grep", "root", "/etc/passwd")

该函数返回杀死进程的信号编号(如果有的话)和执行程序的返回代码。

相关问题