Scenario: A recommendation endpoint experiences significant delays during predictable high-traffic sales events, resulting in poor user experience…

The scenario describes a predictable, recurring high-traffic event (a sales event). The goal is to proactively scale capacity before the event begins to prevent latency. Scheduled scaling is the only proactive scaling policy designed for this exact use case. It allows you to set a schedule to increase the number of instances at a specific time, ensuring the endpoint is prepared for the anticipated load increase. This directly addresses the requirement to prevent delays during predictable peaks by having sufficient capacity ready in advance.

A. Restarting an endpoint causes downtime and does not increase its capacity. This would worsen, not solve, the latency problem during a high-traffic event.

C. Permanently increasing the instance size (vertical scaling) is not a scaling policy. It is a static change that would be inefficient and costly during non-peak hours.

D. Step scaling, like target tracking, is a reactive policy. It adjusts capacity only after a metric (like CPU utilization) crosses a predefined threshold, which means latency may have already occurred.

---

References:

1. AWS SageMaker Developer Guide - "Automatically Scale Amazon SageMaker Models": This official documentation outlines the different scaling policies. It explicitly states, "For predictable load changes, you can use scheduled scaling. For dynamic load changes, you can use step scaling or target tracking scaling policies." This directly supports using scheduled scaling (Option B) for the predictable sales event and refutes using a reactive policy like step scaling (Option D).

Reference Location: AWS SageMaker Developer Guide > SageMaker Hosting > Automatically Scale Amazon SageMaker Models > Scaling policies.

2. Application Auto Scaling User Guide - "Scheduled scaling": This guide, which governs the underlying scaling mechanism for SageMaker, details the purpose of scheduled scaling. It explains, "Scheduled scaling allows you to set your own scaling schedule for predictable load changes." This aligns perfectly with the scenario's need to prepare for a known sales event.

Reference Location: Application Auto Scaling User Guide > Scaling policies > Scheduled scaling.

3. Application Auto Scaling User Guide - "Step scaling and simple scaling policies": This document describes step scaling as a reactive mechanism. It states that step scaling policies "increase or decrease the current capacity...based on a set of scaling adjustments...that vary based on the size of the alarm breach." This confirms that step scaling acts in response to high load, not in anticipation of it, making it unsuitable for the proactive requirement.

Reference Location: Application Auto Scaling User Guide > Scaling policies > Step scaling and simple scaling policies.