multiprocess module with paramiko

I'm trying to use the paramiko python module (1.7.7.1) to execute commands and/or xfer files to a group of remote servers in parallel. One task looks like this:

jobs = []   
for obj in appObjs:
    if obj.stop_app:
        p = m开发者_Go百科ultiprocessing.Process(target=exec_cmd, args=(obj, obj.stop_cmd))
        jobs.append(p)
        print "Starting job %s" % (p)
        p.start()

"obj" contains, among other things, a paramiko SSHClient, transport, and SFTPClient. The appObjs list contains approximately 25 of these objects, and thus 25 connections to 25 different servers.

I get the following error with paramiko's transport.py in the backtrace

raise AssertionError("PID check failed. RNG must be re-initialized after fork(). 
Hint:   Try Random.atfork()")

I patched /usr/lib/python2.6/site-packages/paramiko/transport.py based on the post at https://github.com/newsapps/beeswithmachineguns/issues/17 but it doesn't seem to have helped. I've verified that the transport.py in the path mentioned above is the one being used. The paramiko mailing list appears to have disappeared.

Does this look like a problem in paramiko or am I misunderstanding/misapplying the multiprocessing module? Would anyone be willing to suggest a practical workaround? Many thanks,

---

UPDATE: As @ento notes, the forked ssh package has been merged back into paramiko so the answer below is now irrelevant and you should now being using Paramiko again.

This is a known-problem in Paramiko that has been fixed in a fork of Paramiko (stalled at version 1.7.7.1) that is now just known as the ssh package on pypi (which brings things to version 1.7.11 as of this writing).

Apparently there were problems getting some important patches into the mainline Paramiko and the maintainer was non-responsive, so @bitprophet, the maintainer of Fabric, forked Paramiko under the new package name ssh package on pypi. You can see the specific problem that you mention is discussed here and is one of the reasons he decided to fork it; you can read the gory details if you really want to.

---

As one comment inside the Paramiko issue mentions, the RNG error can be avoided by opening one separate ssh handle per process, Then paramiko will not complain anymore. This sample script demonstrates this (I'm using a Pool instead of Processes):

#!/usr/bin/env python
# -*- coding: utf-8 -*-

import ssh
from multiprocessing import Pool
import getpass

hostnames = [HOST1, HOST2]
user = USERNAME
pw = getpass.getpass("Enter ssh password:")

def processFunc(hostname):
    handle = ssh.SSHClient()
    handle.set_missing_host_key_policy(ssh.AutoAddPolicy())
    handle.connect(hostname, username=user, password=pw)
    print("child")
    stdin, stdout, stderr = handle.exec_command("ls -l /var/log; sleep 5")
    cmdOutput = ""
    while True:
        try:
            cmdOutput += stdout.next()
        except StopIteration:
            break
    print("Got output from host %s:%s" % (hostname, cmdOutput))
    handle.close()

pool = Pool(len(hostnames))
pool.map(processFunc, hostnames, 1)
pool.close()
pool.join()

## If you want to compare speed:
# for hostname in hostnames:
#     processFunc(hostname)