hard20 pts · first answer countsreliabilityincident-responsearchitecture

Outage that will not end

A database blipped for 30 seconds, but the API stayed down for 20 minutes afterwards. Clients retry failed calls immediately, up to 5 times, at every layer. What is the main problem and fix?

A.The database needs more disk and IOPS after the blip; add storage so it can absorb the extra load.B.Raise the retry limit to 20 at every layer, so that each request eventually gets through after recovery.C.The load balancer idle timeout is too long; set it to 1 second so clients give up and fail faster.D.Nested immediate retries create a retry storm; use backoff with jitter, retry at one layer, and circuit breakers.

← First move in a post-deploy outage All replicas landed in one zone →