A company has developed a new ML model. The company requires online model validation on 10% of the traffic before the company fully releases the model in production. The company uses an Amazon…

Question

Options

• AUse production variants to add the new model to the existing SageMaker endpoint. Set the
• BUse production variants to add the new model to the existing SageMaker endpoint. Set the
• CCreate a new SageMaker endpoint. Use production variants to add the new model to the new
• DConfigure the ALB to route 10% of the traffic to the new model at the existing SageMaker

• A
  77% (17)
• B
  5% (1)
• C
  9% (2)
• D
  9% (2)

Explanation

SageMaker Production Variants are purpose-built for this exact use case: they let you host multiple models on a single endpoint and define a InitialVariantWeight to control traffic distribution. Option A uses this native capability on the existing endpoint, requiring only a configuration update - no new infrastructure, no ALB rule changes, and no duplicate endpoint management.

Why the distractors fail:

• B likely sets the traffic weight incorrectly or inverts which variant receives the 10%, making the validation ineffective rather than reducing overhead.
• C creates a new SageMaker endpoint for the new model, which doubles endpoint management costs and complexity - unnecessary when production variants can host both models on the existing endpoint.
• D pushes traffic-splitting logic to the ALB layer, which requires custom routing rules and means SageMaker receives pre-split traffic it can't natively track per variant - more moving parts, more operational overhead.

Memory tip: Think of SageMaker Production Variants as "A/B testing baked in." Whenever an exam question mentions traffic splitting or canary validation between models on SageMaker, Production Variants on the same existing endpoint is almost always the lowest-overhead answer - adding a new endpoint or involving the ALB signals unnecessary complexity.

No community discussion yet for this question.