default random seed not random enough in parallel execution

[klplentific]

Is the below snippet the default random seed?

gen.init_genrand((new Date).getTime() % 1000000000);

If it is it could be unexpected behavior. Because when executed in parallel the chances of matching initial seeds increases like it did in my app where I executed code in parallel.

My issues were gone when I manually set the seeds with a combination of both a date timestamp and Math.random()

I think the principle of least astonishment here would to make the initial seed collision less likely when running code in parallel such as running selenium browser tests.

[jwatte]

If you need high-quality random seeds, I recommend using whatever system-specific random generator is available to you (CryptRandom(), /dev/urand, crypto.getRandomValues(), etc) and initialize using that.

For multiple parallel invocations to share a random seed, each parallel process needs to be created within the same millisecond as the one before it. While that can totally happen, this is not the common case, and even when it happens, it shouldn't break anything.

For the principle of least astonishment, the generator should be seeded by a constant value (such as 1) each time, and the original source of entropy should be entirely left up to the user. So, on the spectrum between:

• initialize with 1
• initialize with millisecond
• initialize with strong system random source the only way to "please everybody" is to provide three separate options, or to require the user to initialize however they want at all times, neither of which seem like an improvement. I don't think Math.random() is a strong system source, either -- it is not seedable by the user, and there are no guarantees that multiple process won't start out with the same initial seed.