AAIA Question of the Day: Circuit Breaker for Serving

Quiz Yazan Ibrahim
AI audit training by expert Yazan Abu Ghosh for auditors with certifications.

AAIA exam practice question — AAIA Question of the Day: Circuit Breaker for Serving

AAIA exam practice question: daily practice for the ISACA Advanced in AI Audit (AAIA) exam — domain: AI Operations.

Question

During a peak traffic event an AI recommendation service experienced rapid autoscaler oscillation and downstream datastore errors, causing increased latency and failed responses. As an AI auditor assessing operational resilience, which control is most effective to prevent cascading failures while preserving user experience?

  • A. Implement a circuit breaker at the model-serving gateway that opens on repeated failures and routes requests to a lightweight fallback model or cached responses.
  • B. Increase the autoscaler's minimum replica count and lengthen scale-up/scale-down cooldowns to maintain higher baseline capacity during spikes.
  • C. Enforce strict per-client rate limits at the API gateway to throttle high-request clients before traffic reaches the model service.
  • D. Provision larger instance types for the serving cluster to increase single-node throughput and reduce the frequency of scaling events.
Show the answer and explanation

Correct answer: A. Implement a circuit breaker at the model-serving gateway that opens on repeated failures and routes requests to a lightweight fallback model or cached responses.

The circuit breaker is the best control because it isolates failing components, stops repeated failing calls from cascading to downstream systems, and enables graceful degradation by routing to a fallback model or cache. This preserves overall availability and reduces error amplification during transient faults. Increasing baseline capacity and cooldowns (option 2) may reduce oscillation but increases cost and does not isolate failures or provide graceful fallback. Per-client rate limiting (option 3) can protect against abusive traffic but may block legitimate users and won't handle backend degradation. Provisioning larger instances (option 4) raises capacity but does not prevent cascading failures or provide fault isolation.

Want more practice?

Prepare for the ISACA Advanced in AI Audit (AAIA) exam with AI Audit & Compliance Framework: Practical Methods & Evaluation.