How to Deploy a Nightly SageMaker Prediction Model?
A company has an ML model that needs to run one time each night to predict stock values. The model input is 3 MB of data that is collected during the current day. The model produces the predictions for the next day. The prediction process takes less than 1 minute to finish running. How should the company deploy the model on Amazon SageMaker to meet these requirements?
Community Insight
The question tests matching SageMaker endpoint type to invocation frequency, payload size, and runtime; the trap is assuming real-time endpoints can scale to zero or that async inference can start with zero instances.
A once-nightly, sub-minute prediction with a 3 MB payload is best served by SageMaker Serverless Inference because it scales to zero between runs and bills only for invocations. The page establishes that option D, with MaxConcurrency 1, meets the requirements while real-time, asynchronous, and multi-model endpoints do not.
The most common wrong choice is a real-time endpoint with autoscaling to zero, but SageMaker real-time endpoints always require at least one provisioned instance.
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
SageMaker Serverless Inference is designed for intermittent or unpredictable traffic and automatically scales to zero when there are no requests, so a model that runs once per night does not need an always-on instance. The 3 MB input and sub-minute execution fit the serverless inference payload and timeout envelope, and MaxConcurrency set to 1 is sufficient for the single nightly invocation while capping concurrency. This directly satisfies the requirement to deploy the model on SageMaker without paying for idle capacity between nightly runs.Why the Other Options Are Wrong
A multi-model serverless endpoint is not the right construct here: multi-model endpoints are aimed at hosting many models behind a provisioned endpoint with model caching, not a single low-frequency nightly job. An asynchronous inference endpoint is intended for large payloads or long-running requests and still requires at least one initial instance; setting InitialInstanceCount to 0 is not a valid way to create the endpoint. A real-time endpoint cannot scale to zero—autoscaling can reduce capacity only to a minimum of one instance—so option C leaves idle cost. Serverless Inference with MaxConcurrency 1 is the only option that matches the short, scheduled workload.Community Comment Notes
Saransundar points out that MaxConcurrency should be set to 1 as the minimum for nightly predictions and pairs it with a 2048 MB memory setting. GiorgioGss captures the deployment logic directly: "why would you provision something in the first place - go serverless." 78e659d adds that one invocation per night with a 3 MB payload and under a minute of processing fits serverless limits and costs nothing between runs.Official Reference
Exam Strategy
Eliminate endpoint types by workload pattern first: nightly scheduled and short-running points to serverless inference (or batch transform), not real-time. Then validate the payload and timeout against the service limits; 3 MB and under one minute fit Serverless Inference, where MaxConcurrency 1 avoids unnecessary concurrency.
Frequently Asked Questions
Why can’t a real-time SageMaker endpoint scale to zero for nightly predictions?
SageMaker real-time endpoints require at least one instance at all times; autoscaling can add instances but cannot reduce the minimum below one. Serverless Inference scales to zero when idle.
Does a 3 MB payload and sub-minute runtime fit SageMaker Serverless Inference?
Yes. The payload and runtime are within Serverless Inference limits, and MaxConcurrency 1 matches the single nightly invocation while allowing scale-to-zero between runs.
Related Analysis
Practice All MLA-C01 Questions
Access 115 questions with complete answers and detailed explanations.
View Full MLA-C01 Practice Test →