如何在多处理中将模块用于“全局”变量？

Question

I am using the common pattern where a module has a number of module-level variables and is imported by other modules, eg module a, b, c... etc import module 'globalv' which has, for examples, module-level variable foo with value 'bar'. So far so good, works as expected in a single-process environment, and one can import the module and/or it's variables. I like this pattern, it makes the code clean and easily readable, and IDEs such as eclipse auto-complete the variable names.

However, in a multiprocessing environment this doesn't work as described in the python documentation because each process has a separate interpreter with distinct address spaces.

Now, I am wondering if perhaps there is commonly accepted way to make this pattern work with multiprocessing ? Perhaps with Manager, Value, or Array ?

My thoughts are to make globalv.foo a Value or Manager instance and pass it to the child processes as and argument; I only use simple, pickleable objects, eg integers, and strings for my "global" variables. The first thing to do when the child process runs is then to replace foo in the child process with the foo passed by the parent process, which is a Value or Manager instance. It should then be the same as the original foo in the parent process and updating foo in the parent or child process should be shared. However, this seems clumsy, any ideas how to do this better be welcome.

Answer 1

You could make use of the shared memory functionality of multiprocessing:

https://docs.python.org/3.9/library/multiprocessing.shared_memory.html