subprocess.check_output（）的问题

Question

I'm having some strange issues using subprocess.check_output() . At first I was just using subprocess.call() and everything was working fine. However when I simply switch out call() for check_output() , I receive a strange error.

Before code (works fine):

def execute(hosts):
''' Using psexec, execute the batch script on the list of hosts '''
successes = []
wd = r'c:\\'
file = r'c:\\script.exe'
for host in hosts:
    res = subprocess.call(shlex.split(r'psexec \\\\%s -e -s -d -w %s %s' % (host,wd,file)))
    if res.... # Want to check the output here
        successes.append(host)
    return successes

After code (doesn't work):

def execute(hosts):
''' Using psexec, execute the batch script on the list of hosts '''
successes = []
wd = r'c:\\'
file = r'c:\\script.exe'
for host in hosts:
    res = subprocess.check_output(shlex.split(r'psexec \\\\%s -e -s -d -w %s %s' % (host,wd,file)))
    if res.... # Want to check the output here
    successes.append(host)
return successes

This gives the error:

I couldnt redirect this because the program hangs here and I can't ctrl-c out. Any ideas why this is happening? What's the difference between subprocess.call() and check_output() that could be causing this?

Here is the additional code including the multiprocessing portion:

PROCESSES = 2
host_sublists_execute = [.... list of hosts ... ]
poolE = multiprocessing.Pool(processes=PROCESSES)
success_executions = poolE.map(execute,host_sublists_execute)
success_executions = [entry for sub in success_executions for entry in sub]
poolE.close()
poolE.join()

Thanks!

Answer 1

You are encountering Python Issue 9400 .

There is a key distinction you have to understand about subprocess.call() vs subprocess.check_output() . subprocess.call() will execute the command you give it, then provide you with the return code . On the other hand, subprocess.check_output() returns the program's output to you in a string, but it tries to do you a favor and check the program's return code and will raise an exception ( subprocess.CalledProcessError ) if the program did not execute successfully (returned a non-zero return code).

When you call pool.map() with a multiprocessing pool, it will try to propagate exceptions in the subprocesses back to main and raise an exception there. Apparently there is an issue with how the subprocess.CalledProcessError exception class is defined, so the pickling fails when the multiprocessing library tries to propagate the exception back to main.

The program you're calling is returning a non-zero return code, which makes subprocess.check_output() raise an exception, and pool.map() can't handle it properly, so you get the TypeError that results from the failed attempt to retrieve the exception.

As a side note, the definition of subprocess.CalledProcessError must be really screwed up, because if I open my 2.7.6 terminal, import subprocess, and manuallly raise the error, I still get the TypeError , so I don't think it's merely a pickling problem.