MLA-C02 practice question: SageMaker Pipelines and CI/CD (2 of 2)
Deployment and Orchestration of ML and AI Workflows A free question from our original MLA-C02 practice exam. Try it, then open the answer to see why each option is right or wrong.
A solar installer retrains its production-output forecasting model every quarter, and it also needs to retrain when production data drifts. Its SageMaker Pipelines workflow already runs training, evaluation and registration, and its existing Model Monitor schedules publish drift metrics to Amazon CloudWatch.
The team wants retraining to start automatically from two sources: a quarterly schedule and detected drift. No one should start the pipeline by hand, and the triggers must stay in managed AWS services.
Which TWO configurations meet these requirements? (Choose two.)
Choose two answers.
AA. Create an Amazon EventBridge rule with a quarterly schedule expression that starts the SageMaker AI pipeline as its target.
BB. Create a CloudWatch alarm on the drift metric that the monitoring schedule publishes, and an EventBridge rule on the alarm's state change that starts the pipeline.
CC. Run a loop on an Amazon EC2 instance that polls the S3 monitoring output every second and starts the pipeline when it finds a new file.
DD. Have the inference code start a pipeline execution with every prediction, so that the model is retrained continuously.
EE. Require a data scientist to start the pipeline from the Studio console after reviewing each monitoring report.
Show the answer and explanation
Answers: A, B
A. Right: Correct. SageMaker Pipelines is a supported EventBridge target, so a scheduled rule can call StartPipelineExecution each quarter with no manual step. The rule needs a role that allows it to start the pipeline. Learn more: Schedule pipeline runs (opens in a new tab)
B. Right: Correct. The monitoring jobs emit drift metrics to CloudWatch, an alarm turns a threshold breach into a state change event, and an EventBridge rule on that event starts the pipeline. Retraining follows drift without any polling code. Learn more: Model Monitor CloudWatch metrics (opens in a new tab)
C. Wrong: Polling every second on a server adds cost and an operated component the stem does not want. Events from EventBridge are the managed trigger for this.
D. Wrong: Starting a pipeline per prediction would retrain far more often than the team needs and would be very costly. The stem asks for scheduled and drift-based triggers.
E. Wrong: Manual starts are excluded by the stem, because the team wants retraining to start automatically. A person running the pipeline by hand is also slower and easier to forget than a managed trigger.
Explanation
EventBridge can start a SageMaker AI pipeline on a schedule and in response to events, such as a CloudWatch alarm on the drift metrics the monitoring schedules publish. That covers both triggers without manual steps or polling. Polling loops, per-prediction retraining and manual starts each break a stated requirement.
Want more on SageMaker Pipelines and CI/CD? The 5 full-length practice sets have 9 more SageMaker Pipelines and CI/CD questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets →