hard20 pts · first answer countsreliability

Retries that make an outage worse

An embedding service had a 30-second blip. Hundreds of clients retried every second with no limit, and the service could not recover for 20 minutes. What should the clients do instead?

A.Use exponential backoff with random jitter, a capped retry count or retry budget, and a circuit breakerB.Retry faster so requests finish soonerC.Never retry anythingD.Open more connections per client

← Streaming counts that run high Error budget for 99.9% →