Threads and stdlib Random

[evanfields]

What’s the state of the Random stdlib and thread safety? I thought I remembered some notes on this from several Julia releases back, but the Random docs have almost nothing on the topic. The only reference to threads is for seed!:

If rng is not specified, it defaults to seeding the state of the shared thread-local generator.

Is rand(rng) threadsafe? What about rand(Random.GLOBAL_RNG)? What is Random.GLOBAL_RNG, since it’s not a MersenneTwister? Thanks!

[jling]

@evanfields

Is rand(rng) threadsafe?

no. basically you should think that when you do rand() in thread, each thread has its own rng. Which makes rand() thread-safe.

[stevengj]

@evanfields

What is Random.GLOBAL_RNG , since it’s not a MersenneTwister ? Thanks!

It’s a special object (of singleton type Random._GLOBAL_RNG) that turns into a per-thread MersenneTwister by calling Random.default_rng().

[stevengj]

@evanfields

the Random docs have almost nothing on the topic.

I made a PR to add some basic info: document thread safety for RNGs by stevengj · Pull Request #40109 · JuliaLang/julia · GitHub

[sylvaticus]

This works as well, I like I can choose to have a (a) random script, (b) deterministic script with individually random functions (same run of the script/model, same output) or (c) deterministic functions (same call with same input, same output), but it requires Julia >= 1.5:

Multithreaded random: different behaviour in Julia >= 1.5 (not just different stream)

Hi, I just realised that my solution below for multithreaded random number generators works only in Julia >= 1.5. In particular the test of “CASE A” fails, i.e. the two repetitions issue the same number. Is there a way to get the same framework working on Julia <= 1.4 ? using Random, Test, Future, Statistics FIXEDRNG = MersenneTwister(123) function generateParallelRngs(rng::AbstractRNG, n::Integer) step = rand(rng,big(10)^20:big(10)^40) # making the step random too ! rngs = Vector{U…