random.choice not random

python, random

Solution

It shouldn't be generating duplicates.

import random
chars = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
def gen():
    return ''.join([random.choice(chars) for x in range(32)])

test = [gen() for i in range(100000)]
print len(test), len(set(test)) # 100000 100000

The chances of duplicates is significant with chars = "ab"; 126 duplicates in 1000000 iterations. It's nonexistant with 62.

That said, this isn't a good way to generate cookies, because session cookies need to be unpredictable, to avoid attacks involving stealing other people's session cookies. The Mersenne Twister is not designed for generating secure random numbers. This is what I do:

import os, hashlib
def gen():
    return hashlib.sha1(os.urandom(512)).hexdigest()

test = [gen() for i in range(100000)]
print len(test), len(set(test))

... which should be very secure (which is to say, difficult to take a string of session cookies and guess other existing session cookies from them).

Problem

I'm using Python 2.5 on Linux, in multiple parallel FCGI processes. I use ``` chars = string.ascii_letters + string.digits cookie = ''.join([random.choice(chars) for x in range(32)]) ``` to generate distinct cookies. Assuming that the RNG is seeded from /dev/urandom, and that the sequence of random numbers comes from the Mersenne twister, I would expect that there is practically zero chance of collision. However, I do see regular collisions, even though only a few (<100) users are logged in at any time. Why are the random numbers not more random?

Original source