We were able to implement our own incantation of this and use it live with over 120k users live. We peaked at around 15% cpu when large batches of users were joining, and leveled out at around ~7% cpu.
I’ve prepared a gist of the main parts of the code for the benefit of future readers.
I still think there are improvements to be made here. One downside is if the node that has the global gen_server running crashes for some reason, we won’t start up a new one.
If anyone ends up using this code and making improvements, please report back!
