Can we do better in Python calling multi-thread Julia function?

[gRox167]

Hi, I mainly use Julia to write computational intensive function to mitigate the overhead of python, and use python just as a script to call Julia function. I am also writing a package that my lab member can use in python (that's why I use Python to call Julia instead of use Julia to call Python). The goal that I rewrite original python code in julia is to accelerate original python code, with julia's fast and simple UX for thread based parallel computing.
For example,

function test()
    Threads.@threads for i in 1:20
        a = rand(1000,1000,1000).^2
    end
end

I find out that in python calling julia function with Threads.@threads macro inside will lead to segfault without turning of GC, even this function does not include any python object. Currently the best I can get is PythonCall.GC.disable() before and PythonCall.GC.enable() after computation.

function test()
    PythonCall.GC.enable()
    Threads.@threads for i in 1:20
        a = rand(1000,1000,1000).^2
    end
    PythonCall.GC.disable()
end

Sometimes, disable GC will cause memory overflow, we need to manually collect the unused resource, which is a tedious work. This will lose the simplicity of julia's thread macro and introduce unnecessary cognitive burden to the user, I came up with this discussion to talk about a better way to handle for python to call julia multi-thread function.

I am not a professional computer science student, so there may be a lot of limitation for my understanding. I am not very familiar with the idea of different parallel model and different garbage collection solution. If I made any mistake please feel free to point it out.