Python:How os.fork() works?

I am learn multi processing in python. I tried multiprocessing and after I read the source code of multiprocessing module, i found it use os.fork(),so i write some code to test os.fok(), but i am stucked. My code is as following:

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

import os
import time

for i in range(2):
    print '**********%d***********' % i
    pid = os.fork()
    print "Pid %d" % pid

I think the each print will be executed two times but they execute three times.I can't understand how this work? I read this Need to know how fork works?
From what this article says it also will be executed twice,so i am so stucked......

回答1:

To answer the question directly, os.fork() works by calling the underlying OS function fork().

But you are surely interested in what this does. Well this creates another process which will resume at exactly the same place as this one. So within the first loop run, you get a fork after which you have two processes, the "original one" (which gets a pid value of the PID of the child process) and the forked one (which gets a pid value of 0).

They both print their pid value and go on with the 2nd loop run, which they both print. Then they both fork, leaving you with 4 processes which all print their respective pid values. Two of them should be 0, the other two should be the PIDs of the child they just created.

Changing the code to

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

import os
import time

for i in range(2):
    print '**********%d***********' % i
    pid = os.fork()
    if pid == 0:
        # We are in the child process.
        print "%d (child) just was created by %d." % (os.getpid(), os.getppid())
    else:
        # We are in the parent process.
        print "%d (parent) just created %d." % (os.getpid(), pid)

you will see better what happens: each process will tell you its own PID and what happened on the fork.

回答2:

First of all, remove that print '******...' line. It just confuses everyone. Instead, let's try this code...

import os
import time

for i in range(2):
    print "I'm about to be a dad!"
    time.sleep(5)
    pid = os.fork()
    if pid == 0:
        print "I'm {}, a newborn that knows to write to the terminal!".format(os.getpid())
    else:
        print "I'm the dad of {}, and he knows to use the terminal!".format(pid)
        os.waitpid(pid)

Okay, first of all, what is "fork"? Fork is a feature of modern and standard-compliant operating systems (except of M$ Windows: that joke of an OS is all but modern and standard-compliant) that allows a process (a.k.a: "program", and that includes the Python interpreter!) to literally make an exact duplicate of itself, effectively creating a new process (another instance of the "program"). Once that magic is done, both processes are independent. Changing anything in one of them does not affect the other one.

The process responsible for spelling out this dark and ancient incantation is known as the parent process. The soulless result of this inmoral abomination towards life itself is known as the child process.

As shall be obvious to all, including those for which it isn't, you can become a member of that select group of programmers who have sold their soul by means of os.fork(). This function performs a fork operation, and thus results in a second process being created out of thin air.

Now, what does this function return, or more importantly, how does it even return? If you want not to become insane, please don't go and read the Linux kernel's /kernel/fork.c file! Once the kernel does what we know it has to do, but we don't want to accept it, os.fork() returns in the two processes! Yes, even the call stack is copied on!

So, if their are exact copies, how does one difference between parent and child? Simple. If the result of os.fork() is zero, then you're working in the child. Otherwise, you're working in the parent, and the return value is the PID (Process IDentifier) of the child. Anyway, the child can get its own PID from os.getpid(), no?

Now, taking this into account, and the fact that doing fork() inside a loop is the recipe for mess, this is what happens. Let's call the original process the "master" process...

Master: i = 0, forks into child-#1-of-master
- Child-#1-of-master: i = 1 forks into child-#1-of-child-#1-of-master
- Child-#1-of-child-#1-of-master: for loop over, exits
- Child-#1-of-master: for loop over, exits
Master: i = 1, forks into child-#2-of-master
- Child-#2-of-master: i = 1 forks into child-#1-of-child-#2-of-master
- Child-#1-of-child-#2-of-master: for loop over, exits
- Child-#2-of-master: for loop over, exits
Master: for loop over, exits

As you can see, there are a total of 6 parent/child prints coming from 4 unique processes, resulting in 6 lines of output, something like...

I'm the dad of 12120, and he knows to use the terminal!

I'm 12120, a newborn that knows to write to the terminal!

I'm the dad of 12121, and he knows to use the terminal!

I'm 12121, a newborn that knows to write to the terminal!

I'm the dad of 12122, and he knows to use the terminal!

I'm 12122, a newborn that knows to write to the terminal!

But that's just arbitrary, it could have output this out instead...

I'm 12120, a newborn that knows to write to the terminal!

I'm the dad of 12120, and he knows to use the terminal!

I'm 12121, a newborn that knows to write to the terminal!

I'm the dad of 12121, and he knows to use the terminal!

I'm 12122, a newborn that knows to write to the terminal!

I'm the dad of 12122, and he knows to use the terminal!

Or anything other than that. The OS (and your motherboard's funky clocks) is the sole responsible for the order in which processes get timeslices, so go blame on Torvalds (and expect no self-steem when back) if you dislike how the kernel manages to organize your processes ;).

I hope this has led some light on you!