MLA-C01 Deployment and Orchestration of ML Workflows Practice Question

A data science team is using AWS Step Functions to orchestrate a machine learning workflow that includes a SageMaker training job followed by a model deployment. They want to ensure that if the training job fails, the workflow retries up to three times with exponential backoff before sending a notification to an Amazon SNS topic. Which Step Functions feature should they use to implement this?

⚠ Common exam trap

The trap here is using a Choice state or Map state to implement retries, which lacks native exponential backoff and complicates error handling compared to Retry and Catch.

Answer choices

Why each option matters

Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.

Correct answer & explanation

✓

Retry and Catch fields on the training task state, with a retry policy specifying MaxAttempts and BackoffRate, and a Catch field that transitions to an SNS publish state.

Step Functions' Retry and Catch fields on a state provide built-in error handling. By configuring Retry with MaxAttempts and BackoffRate, the training task will be retried with exponential backoff. The Catch field can then route to an SNS publish state after all retries fail, ensuring notification only on persistent failure.

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

✗
A Map state that iterates over a list of retry attempts, invoking the training job each time until it succeeds.
Why it's wrong here
A Map state is for iterating over a collection, not for retrying a single task. Using a Map state to retry would require pre-defining a list and would not provide exponential backoff or conditional transitions based on failure. It is not the intended mechanism for retries.
✗
A Choice state that checks the training job status and loops back to the training state if it failed, with a counter to limit attempts.
Why it's wrong here
While a Choice state can create a loop, implementing retries with exponential backoff and a maximum attempt limit would require complex state management and custom logic. Step Functions provides native Retry and Catch fields that handle this more elegantly and with less overhead.
✓
Retry and Catch fields on the training task state, with a retry policy specifying MaxAttempts and BackoffRate, and a Catch field that transitions to an SNS publish state.
Why this is correct
Step Functions allows you to define Retry and Catch on individual states. The Retry field can specify MaxAttempts, IntervalSeconds, and BackoffRate to implement exponential backoff. The Catch field can transition to a fallback state, such as an SNS publish task, when retries are exhausted. This directly meets the requirement.
✗
A Parallel state that runs the training job and an SNS notification simultaneously, with a retry policy on the training job.
Why it's wrong here
A Parallel state is used to execute branches concurrently, not to handle retries or error handling. While you could run the training job and SNS in parallel, the SNS notification would be sent regardless of training success, which is not desired. The requirement is to notify only after retries fail.

Go deeper

MLA-C01 Deployment and Orchestration of ML Workflows Practice Question

Why each option matters

Option-by-option breakdown

Related to this question