When it comes to rate limiting for microservices, should you do it per-service or globally? That's the big question.

hey folks, ive been thinkin bout puttin rate limitin on our microservices architecture. the main q is if its better to set limits on each service separately (like the user service limits its own stuff) or do a global rate limiter at the edge or api gateway. each way seems to have good and bad points for how complicated it is, if its fair, and stoppin cascading failures. what have u all found works best in real life? any gotchas to watch out for with either way