I don't exactly remember how i implemented it, but it basically did a single call to redis to count the request for the IP and check the limits.
Another usecase where the more advanced data types & operations of redis are usefull, is for job queues, since you can atomically move a job from the 'queue' to the 'processing' list, thus preventing loosing jobs if the processor crashes after pulling it orso. But we do run all those on persisted redis stores, for safety :)
And if i would do it all again, i'd probably just use postgres for anything i want to keep when things crash. Redis just kinda lives between a 'real' database and a pure volatile kv-cache