You work for an organization that manages an ecommerce site. Your application is deployed behind an external Application Load Balancer. You need to test a new product recommendation algorithm. You pla

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
  80% (33)
• B
  2% (1)
• C
  12% (5)
• D
  5% (2)

Explanation

Traffic splitting with percentage weights is the standard mechanism for A/B testing behind a load balancer. You configure two backend service versions and assign weights (e.g., 50%/50% or 90%/10%), and the load balancer randomly routes users to each version. This creates a statistically valid, randomized experiment suitable for measuring the impact of the new recommendation algorithm on sales. Option B (feature flag on a single instance) is not randomized and limits exposure to one server, which is not representative. Option C (traffic mirroring) duplicates traffic to the new version but the mirrored requests are fire-and-forget-they don't affect real users and cannot measure sales impact. Option D (HTTP header-based routing) is deterministic and non-random, requiring specific headers rather than random assignment.

No community discussion yet for this question.