Capping the Comeback

Your service caps every call at 3 attempts. During a partial brownout, the dependency starts failing about 40% of calls, and the retries push its total inbound load well past what it can serve, so it never recovers. Which control most directly stops that?

A higher attempt cap of 5 with longer backoff, so a request keeps trying until the dependency answers

A retry budget that allows retries only up to a small fraction of recent successful calls and drops the rest

Full jitter on the backoff so the retries are spread evenly instead of arriving in synchronized waves

Shorter per-call timeouts so failing calls release the caller's threads and connections much sooner