GCP Professional Cloud Architect Practice Question

A media-streaming firm will launch a live-sports platform on GKE in two regions behind an external HTTP(S) Load Balancer. Normal events peak at 5 000 RPS; marketing predicts celebrity promotions could hit 10 000 RPS. Pre-production tests must (1) confirm the service meets its 200-ms 95th-percentile latency SLO at 5 000 RPS and (2) reveal the system's failure and recovery behavior under extreme overload. Which testing strategy best meets both objectives according to Google Cloud reliability practices?

Conduct a 24-hour soak test at 5 000 RPS to detect memory leaks, followed by routine synthetic uptime checks; skip any higher-than-normal load generation to protect production-like resources.
Run a staged load test that ramps to 5 000 RPS and sustains it for an hour, then launch a separate stress test that doubles traffic every five minutes until 10 000 RPS is reached or error rates spike, observing recovery once load is removed.
Rely solely on chaos engineering: periodically terminate random GKE pods during normal traffic without additional high-volume traffic generation, trusting horizontal pod autoscaling to prove resilience.
Execute a single stress test by immediately sending 10 000 RPS to the service and abort the test once latency exceeds 200 ms.

Report Issue

Answer Description

Load testing validates that a system meets performance objectives under expected traffic. Ramping up to the known 5 000-RPS peak and sustaining that rate while measuring 95th-percentile latency confirms the platform can satisfy its 200-ms SLO during normal events. Stress testing intentionally pushes traffic far beyond the anticipated maximum-here, increasing toward 10 000 RPS-so engineers can observe how the platform degrades, verify that autoscaling and circuit-breaker policies activate correctly, and ensure the service recovers once load is removed. This two-phase approach fulfills both stated goals, whereas the alternatives omit either the normal-load validation or the controlled overload analysis, or apply unrelated test types.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.