First of all, thanks for sharing this very detailed improvement with us! I really appreciate this content.

I’m only wondering why you’re not using find_in_bacthes here too! I mean, of course allocating a array of integer only is too less expensive which one full of User class objects(with a thousand of dependents objects), but this can be a way to keep this scaling when you have thousand of millions records.

Something like this: https://gist.github.com/siannopollo/03d646eb7525f7fce678#file-pluck_in_batches-rb