MLA-C01 Practice Question 57: Pipeline Step Caching
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 6 · question 7 of 10
A SageMaker Pipelines workflow has four steps: a processing step that takes two hours, a training step that takes three hours, an evaluation step, and a model registration step. Engineers are now iterating on the evaluation script and rerun the pipeline several times a day. Between these runs the processing and training steps have identical inputs, arguments, and code, but every execution still spends five hours recomputing them before evaluation starts.
What is the most direct way to stop rerunning the unchanged steps?
Show answer and explanation
Answer:
Step caching in SageMaker Pipelines reuses the output of an earlier successful run of a step, instead of recomputing it, when the step has the same configuration and inputs. Pipelines compares the step's cache key attributes with previous runs and skips the step when they match. You turn it on per step with a cache configuration that can also set an expiry. Processing and training then return almost immediately, and only the changed evaluation step actually runs (option B).
- A is wrong because warm pools keep instances available to cut *startup* time for the next training job. The job still reruns all three hours of training. - C is wrong because bigger or more instances only shorten work that should not happen at all, and they raise the cost of every run. - D is wrong because the first pipeline still recomputes processing and training on every run before it triggers evaluation, and the workflow now has two pipelines to maintain.
Reference: Caching pipeline steps