Online Validation of a New Model on Ten Percent of Traffic with SageMaker Production Variant Weight
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 SageMaker endpoint behind an Application Load Balancer (ALB) to serve the model. Which solution will set up the required online validation with the LEAST operational overhead?
Community Votes
100% of anonymous learners picked answer A. Votes are pick records left by other test-takers — they are not the verified answer.
Community Insight
SageMaker production variants split traffic across multiple models on a single existing endpoint by variant weight, so setting the new model's weight to 0.1 sends exactly 10% of invocations to it with no new endpoint and no load balancer change.
A company requires online model validation on 10% of traffic for a new ML model before fully releasing it in production, and it already serves a SageMaker endpoint behind an Application Load Balancer. The setup must achieve the online validation with the least operational overhead.
Creating a second SageMaker endpoint and splitting variants there, which duplicates infrastructure to achieve something the existing endpoint's variant weights already provide. Configuring the ALB to route traffic and monitoring with CloudTrail adds load balancer management and uses the wrong telemetry.
Community Discussion (3 comments)
Comments & Corrections
No comments yet — spotted an error or have a note? Share it below.
Expert Analysis
Why the Answer Is Correct
The requirement is online validation of a new model on 10% of traffic before full release, with the least operational overhead, and the company already has a SageMaker endpoint serving the current model. SageMaker production variants allow multiple models to be hosted on a single endpoint with traffic split by variant weight, so adding the new model as a variant with a weight of 0.1 routes exactly ten percent of invocations to it while the existing model keeps the rest, and the allocation happens inside the endpoint with no load balancer change. Invocation counts are then observed with CloudWatch metrics. The vote was unanimous at 100 for A. GiorgioGss provided the exact ProductionVariants configuration with the existing model at 0.9 and the new model at 0.1, aws_Tpozit noted that production variants allow traffic splitting across multiple models on a single endpoint, and Saransundar cited the SageMaker A/B testing documentation.Why the Other Options Are Wrong
Setting the new model's variant weight to 1 (B) sends all traffic to the new model, which is a full release rather than a 10% online validation, so it does not meet the requirement at all. Creating a new SageMaker endpoint and adding production variants there (C) achieves the split but provisions a second endpoint to duplicate capacity that the existing endpoint already has, which is added infrastructure and therefore higher operational overhead. Configuring the ALB to route 10% of traffic and monitoring invocations with CloudTrail (D) moves the traffic split up into the load balancer, which adds load balancer configuration to manage, and it monitors with CloudTrail, which records API events rather than providing the per-variant invocation metrics needed to confirm the split.Community Comment Notes
The community was unanimous at 100 for A, and GiorgioGss gave the most concrete answer by supplying the actual ProductionVariants JSON showing the existing model at 0.9 and the new model at 0.1, which demonstrates both the 90/10 split and that it is a single endpoint. Saransundar cited the official A/B testing documentation, and aws_Tpozit identified the keyword as validating on 10% of traffic with minimal overhead, which is precisely what a low variant weight on an existing endpoint delivers.Official Reference
Related Analysis
Practice All MLA-C01 Questions
Access 115 questions with complete answers and detailed explanations.
View Full MLA-C01 Practice Test →