How are `remote functions` and `remote actors` made to run in a distributed manner under the hood in the ray library?

Question

I am interested in understanding how remote functions and remote actors are made to run in a distributed manner under the hood in the ray library. I have traced the code beginning from ray.remote to the following lines thus far:

• For functions: https://github.com/ray-project/ray/blob/6233cef22cc3c62034b4a40923d1eaebdfef883f/python/ray/worker.py#L1710

• For class: https://github.com/ray-project/ray/blob/6233cef22cc3c62034b4a40923d1eaebdfef883f/python/ray/worker.py#L1732

Beyond the above point, could you provide simple intuitive step-by-step explanation of how a function and a class is converted and ran in a distributed manner?

Thank you for your help.

Answer 1

Please checkout the whitepaper for more internal details; https://docs.google.com/document/d/1lAy0Owi-vPz2jEqBSaHNQcy2IBSDEHyXNOQZlGuj93c/preview