Cloud Native Operations

English

한국어

English

한국어

Appearance

SageMaker AI and MLflow Execution Quiz ​

Multiple Choice Questions ​

  1. Which resource is used for new managed MLflow deployments?
    • A) Only the legacy Tracking Server
    • B) SageMaker MLflow App
    • C) A public EKS LoadBalancer
    • D) Only a local SQLite file
Show Answer

Answer: B

The managed path uses the current SageMaker MLflow App API.

  1. What is required before a full run?
    • A) A completed smoke run and raw-PII logging scan
    • B) Removal of project membership
    • C) Immediate maximum-step execution
    • D) Disabled teardown
Show Answer

Answer: A

Review smoke completion, logging safety, aggregate results, and adapter inventory first.

  1. What is the EKS Job failure/deadline contract?
    • A) Unlimited retries and no deadline
    • B) backoffLimit: 0 and activeDeadlineSeconds: 10800
    • C) Ten retries and one hour
    • D) A permanent Deployment
Show Answer

Answer: B

These are retry and Job deadline settings. Controller or node failures can delay reclamation; they do not guarantee all resources disappear within three hours or cap total cost.

  1. When must cleanup verification fail?
    • A) No resources remain
    • B) Resources remain or errors prevent confirming their state
    • C) Smoke succeeds
    • D) Dataset hashes match
Show Answer

Answer: B

Zero remaining is insufficient when queries failed. Preserve unknown states and refuse name-only deletion from old inventories without ownership evidence.

  1. What was actually exercised on September 1?
    • A) A full SageMaker Training Job
    • B) A full EKS GPU Job
    • C) Provisioning and cleanup paths
    • D) Tuned-model evaluation
Show Answer

Answer: C

Both GPU Jobs stopped before submission.

  1. Why is GPU execution blocked as of September 12, 2026?
    • A) The pinned PyTorch 2.8 DLC reached end of patch on August 6
    • B) Source bundling is billable
    • C) Every MLflow App was deleted
    • D) Local unit tests require a GPU
Show Answer

Answer: A

Validate a supported DLC, torch, dependency cohort, and GPU smoke run. Removing the date check alone is not an upgrade.

  1. What does the SageMaker launcher do without --execute?
    • A) Submit the full Job immediately
    • B) Write the request JSON only
    • C) Delete existing Jobs
    • D) Guarantee a cost estimate
Show Answer

Answer: B

Request preview is the default. Actual submission requires the runtime support check and separate review.

  1. Is saving EKS MLflow metric/parameter JSON enough?
    • A) Yes, it automatically includes adapter weights
    • B) Download final adapters and aggregate artifacts, then verify hashes
    • C) emptyDir survives cluster deletion
    • D) A log tail can reconstruct the model
Show Answer

Answer: B

Metadata export is different from preserving files. A cluster retained after export failure also needs cost awareness and subsequent cleanup.

Return to learning materials