A retailer's recommendation endpoint must keep p99 latency under 50 ms during a flash sale that could bring ten times normal load. Which configuration best meets this requirement?
Show answer and explanation
Answer: C
Autoscaling with a sensible minimum, a raised maximum, and a CPU-based target lets the endpoint add capacity as load rises and keep latency within budget, without relying on fixed guesses. A fixed pool cannot absorb a tenfold surge, one large replica is a single point of failure, and manual scaling reacts too slowly and is prone to error.
Why the other options are wrong
- A. Two fixed replicas cannot absorb a tenfold increase, so queues build up and tail latency climbs far beyond the 50 ms budget.
- B. One bigger replica still has a throughput ceiling and gives no redundancy, so a failure or saturation takes the whole endpoint down.
- D. Hand scaling depends on someone acting before the surge and on accurate forecasts, so it reacts late and fails whenever the sale starts early.