Short answer: mostly Stop cascading failures in a complex distributed system.
I have no meaningful alternative answer so I need a circuit breaker?
This question is relevant only if your server serves only one REST endpoint (and one HTTP verb). But almost always, it is not. Even microservices will have a combination of several endpoints + several http verbs. You don’t want one endpoint to hang on slow microservice threads and heaps after the threads continue to wait there and eventually drop your entire application.
Read the official documentation
What is histrix? --- Hystrix is designed to perform the following actions:
"Fallback and gracefully degrade when possible" - , hystrix.