使用subprocess在Python中逐行链式读写管道

Question

我有一段代码，看起来可以在Python中通过子进程将多个命令串联起来，并且可以逐行读取和写入数据（而不是提前使用communicate()）。这段代码调用了一个Unix命令（mycmd），读取它的输出，然后把这些输出写入另一个Unix命令（next_cmd）的输入，并把最后一个命令的输出重定向到一个文件中。

    # some unix command that uses a pipe: command "a"
    # writes to stdout and "b" reads it and writes to stdout
    mycmd = "a | b" 
    mycmd_proc = subprocess.Popen(mycmd, shell=True,
                                  stdin=sys.stdin,
                                  stdout=subprocess.PIPE,
                                  stderr=subprocess.PIPE)
    # nextCmd reads from stdin, and I'm passing it mycmd's output
    next_cmd = "nextCmd -stdin"
    output_file = open(output_filename, "w")
    next_proc = subprocess.Popen(next_cmd, shell=True,
                                  stdin=subprocess.PIPE,
                                  stdout=output_file)
    for line in iter(mycmd.stdout.readline, ''):
        # do something with line
        # ...
        # write it to next command
        next_proc.stdin.write(line)
    ### If I wanted to call another command here that passes next_proc output
    ### line by line to another command, would I need
    ### to call next_proc.communicate() first?
    next_proc.communicate()
    output_file.close()

这个方法看起来有效，并且它只在命令结束时调用communicate()。

我想扩展这段代码，添加另一个命令，这样你就可以做到：

mycmd1 | mycmd2 | mycmd3 > some_file

意思是：逐行读取来自Python的mycmd1的输出，处理这一行，然后把它传给mycmd2，读取mycmd2的输出，逐行处理它，再把它传给mycmd3，最后mycmd3的输出会放到some_file中。这可能吗？还是说这会导致死锁、阻塞或者缓冲区未刷新？请注意，我并不是简单地将三个Unix命令串联在一起，因为我想在中间用Python进行干预，并逐行处理每个命令的输出，然后再传给下一个命令。

我想避免调用communicate并将所有输出加载到内存中——我希望逐行解析输出。谢谢。

Answer 1

这个方法可以处理任意数量的命令：

import sys
import subprocess

def processFirst(out):
    return out

def processSecond(out):
    return out

def processThird(out):
    return out

commands = [("a|b", processFirst), ("nextCmd -stdin", processSecond), ("thirdCmd", processThird)]

previous_output = None
for cmd,process_func in commands:
    if previous_output is None:
        stdin = sys.stdin
    else:
        stdin = subprocess.PIPE
    proc = subprocess.Popen(cmd, shell=True,
                            stdin = stdin,
                            stdout = subprocess.PIPE)
    if previous_output is not None:
        proc.stdin.write(previous_output)

    out,err = proc.communicate()
    out = process_func(out)
    previous_output = out

你只需要把想要运行的命令添加到命令列表中，并指定一个函数来处理它的输出。最后一个命令的输出会在循环结束时存放在 previous_output 里。

为了避免出现死锁、缓冲等问题，你可以使用 proc.communicate() 来确保每个命令都能完整运行，这样它会返回输出（而不是像你例子中那样直接读取）。然后把这个输出传给下一个命令，再让它运行，依此类推。

补充一下：我刚注意到你不想一开始就使用 communicate()，而是想逐行反应。我会稍后修改我的回答来解决这个问题。

这个回答提供了一个如何使用 select.select() 从管道逐行读取而不阻塞的例子。

下面是一个针对你具体情况的例子：

import sys
import subprocess
import select
import os

class LineReader(object):
    def __init__(self, fd, process_func):
        self._fd = fd
        self._buf = ''
        self._process_func = process_func
        self.next_proc = None

    def fileno(self):
        return self._fd

    def readlines(self):
        data = os.read(self._fd, 4096)
        if not data:
            # EOF
            if self.next_proc is not None:
                self.next_proc.stdin.close()
            return None
        self._buf += data
        if '\n' not in data:
            return []
        tmp = self._buf.split('\n')
        tmp_lines, self._buf = tmp[:-1], tmp[-1]
        lines = []
        for line in tmp_lines:
            lines.append(self._process_func(line))
            if self.next_proc is not None:
                self.next_proc.stdin.write("%s\n" % lines[-1])

        return lines

def processFirst(line):
    return line

def processSecond(line):
    return line

def processThird(line):
    return line

commands = [("a|b", processFirst), ("nextCmd -stdin", processSecond), ("thirdCmd", processThird)]

readers = []
previous_reader = None
for cmd,process_func in commands:
    if previous_reader is None:
        stdin = sys.stdin
    else:
        stdin = subprocess.PIPE
    proc = subprocess.Popen(cmd, shell=True,
                            stdin = stdin,
                            stdout = subprocess.PIPE)

    if previous_reader is not None:
        previous_reader.next_proc = proc

    previous_reader = LineReader(proc.stdout.fileno(), process_func)
    readers.append(previous_reader)


while readers:
    ready,_,_  = select.select(readers, [], [], 10.0)
    for stream in ready:
        lines = stream.readlines()
        if lines is None:
            readers.remove(stream)