writing large amount of data to stdin

问题:

I am writing a large amount of data to stdin.

How do i ensure that it is not blocking?

p=subprocess.Popen([path],stdout=subprocess.PIPE,stdin=subprocess.PIPE)
p.stdin.write('A very very very large amount of data')
p.stdin.flush()
output = p.stdout.readline()

It seems to hang at p.stdin.write() after i read a large string and write to it.

I have a large corpus of files which will be written to stdin sequentially(>1k files)

So what happens is that i am running a loop

#this loop is repeated for all the files
for stri in lines:
p=subprocess.Popen([path],stdout=subprocess.PIPE,stdin=subprocess.PIPE)
p.stdin.write(stri)
output = p.stdout.readline()
#do some processing

It somehow hangs at file no. 400. The file is a large file with long strings.

I do suspect its a blocking issue.

This only happens if i iterate from 0 to 1000. However, if i were to start from file 400, the error would not happen

回答1:

To avoid the deadlock in a portable way, write to the child in a separate thread:

#!/usr/bin/env python
from subprocess import Popen, PIPE
from threading import Thread

def pump_input(pipe, lines):
    with pipe:
        for line in lines:
            pipe.write(line)

p = Popen(path, stdin=PIPE, stdout=PIPE, bufsize=1)
Thread(target=pump_input, args=[p.stdin, lines]).start()
with p.stdout:
    for line in iter(p.stdout.readline, b''): # read output
        print line,
p.wait()

See Python: read streaming input from subprocess.communicate()

回答2:

You may have to use Popen.communicate().

If you write a large amount of data to the stdin and during this the child process generates output to stdout then it may become a problem that the stdout buffer of the child becomes full before processing all of your stdin data. The child process blocks on a write to stdout (because you are not reading it) and you are blocked on writing the stdin.

Popen.communicate() can be used to write stdin and read stdout/stderr at the same time to avoid the previous problem.

Note: Popen.communicate() is suitable only when the input and output data can fit to your memory (they are not too large).

Update: If you decide to hack around with threads here is an example parent and child process implementation that you can tailor to suit your needs:

parent.py:

#!/usr/bin/env python2
import os
import sys
import subprocess
import threading
import Queue


class MyStreamingSubprocess(object):
    def __init__(self, *argv):
        self.process = subprocess.Popen(argv, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
        self.stdin_queue = Queue.Queue()
        self.stdout_queue = Queue.Queue()
        self.stdin_thread = threading.Thread(target=self._stdin_writer_thread)
        self.stdout_thread = threading.Thread(target=self._stdout_reader_thread)
        self.stdin_thread.start()
        self.stdout_thread.start()

    def process_item(self, item):
        self.stdin_queue.put(item)
        return self.stdout_queue.get()

    def terminate(self):
        self.stdin_queue.put(None)
        self.process.terminate()
        self.stdin_thread.join()
        self.stdout_thread.join()
        return self.process.wait()

    def _stdin_writer_thread(self):
        while 1:
            item = self.stdin_queue.get()
            if item is None:
                # signaling the child process that the end of the
                # input has been reached: some console progs handle
                # the case when reading from stdin returns empty string
                self.process.stdin.close()
                break
            try:
                self.process.stdin.write(item)
            except IOError:
                # making sure that the current self.process_item()
                # call doesn't deadlock
                self.stdout_queue.put(None)
                break

    def _stdout_reader_thread(self):
        while 1:
            try:
                output = self.process.stdout.readline()
            except IOError:
                output = None
            self.stdout_queue.put(output)
            # output is empty string if the process has
            # finished or None if an IOError occurred
            if not output:
                break


if __name__ == '__main__':
    child_script_path = os.path.join(os.path.dirname(__file__), 'child.py')
    process = MyStreamingSubprocess(sys.executable, '-u', child_script_path)
    try:
        while 1:
            item = raw_input('Enter an item to process (leave empty and press ENTER to exit): ')
            if not item:
                break
            result = process.process_item(item + '\n')
            if result:
                print('Result: ' + result)
            else:
                print('Error processing item! Exiting.')
                break
    finally:
        print('Terminating child process...')
        process.terminate()
        print('Finished.')

child.py:

#!/usr/bin/env python2
import sys

while 1:
    item = sys.stdin.readline()
    sys.stdout.write('Processed: ' + item)

Note: IOError is processed on the reader/writer threads to handle the cases where the child process exits/crashes/killed.