You work for an organization that manages an ecommerce site. Your application is deployed behind a global HTTP(S) load balancer. You need to test a new product recommendation algorithm. You plan to us

Question

Options

• ASplit traffic between versions using weights.
• BEnable the new recommendation feature flag on a single instance.
• CMirror traffic to the new version of your application.
• DUse HTTP header-based routing.

• A
  77% (23)
• B
  7% (2)
• C
  13% (4)
• D
  3% (1)

Explanation

Traffic splitting by weight is the standard mechanism for A/B testing in Cloud Run or behind a load balancer. By assigning a percentage of traffic (e.g., 50%) to the new recommendation algorithm version and the remainder to the existing version, users are randomly routed to either version, allowing you to statistically compare sales outcomes. This is randomized by nature since routing is determined by the load balancer without bias. Option B (feature flag on a single instance) is not randomized and only affects a subset of a single instance. Option C (traffic mirroring) sends a copy of traffic to the new version but the mirrored traffic does not serve real responses, so it cannot measure sales impact. Option D (HTTP header-based routing) targets specific users based on request headers, which is not random.

No community discussion yet for this question.