MLA-C01 Practice Question: A data engineer is designing a feature…

Answer analysis

Option-by-option breakdown

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

• ✗

  Create a feature group with only an online store

  Why it's wrong here

  An online-only feature group retains just the latest feature values for low-latency reads, so point-in-time historical retrieval for training is impossible. It is tempting because it satisfies the millisecond inference requirement, and would be correct if only real-time serving were needed.

• ✗

  Create separate feature groups — one for online and one for offline — and manage data synchronization manually

  Why it's wrong here

  SageMaker Feature Store already writes each record to both stores from one feature group; splitting them forces manual synchronisation and risks divergence between training and serving values. It is tempting because it appears to separate concerns, and would be correct only where the two stores genuinely hold different data.

• ✓

  Create a feature group with both online and offline stores enabled

  Why this is correct

  Enabling both stores gives the feature group a low-latency online store for millisecond real-time inference and an offline store retaining historical values with event times for point-in-time batch training retrieval. This single configuration satisfies both the latency and historical-access constraints stated in the stem.

• ✗

  Store features only in the offline store and use a separate low-latency cache like ElastiCache

  Why it's wrong here

  An offline store alone serves historical queries, not millisecond reads, and ElastiCache holds no point-in-time versioning, so training would lose temporal correctness. It is tempting because caching delivers low latency, and would be correct for serving precomputed values where historical accuracy is irrelevant.

Go deeper

Related to this question