You run a LAMP-based user-feedback portal in us-central1 across two zones with autoscaled managed instance groups except the DB. It currently serves a small group with 99.99% availability. Before opening to all users, which resiliency test strategy ensures the system maintains SLA under higher, unauthenticated load?

The correct strategy involves creating synthetic random user input and replaying this load until autoscaling is triggered on at least one layer. This simulates realistic, unauthenticated user behavior without relying on existing user data, which might not reflect future usage patterns. Introducing chaos by terminating random resources across both zones (e.g., VMs, network interfaces) is crucial for testing the system's resilience and fault tolerance under stress, ensuring it maintains SLA even when components fail. Incorrect options: Replaying captured user input and terminating an entire zone tests resilience but might not fully simulate the unauthenticated load or the impact of random, smaller failures. Gradually exposing to a larger user group is a slow rollout, not a stress test for resilience under high load and failure conditions. Replaying captured user input until 80% utilization and pre-provisioning resources doesn't test the autoscaling mechanism's effectiveness or the system's resilience to failures.