Amazon AWS Certified Machine Learning Engineer - Associate MLA-C01 Practice Test 99
Question 980
While reproducing the alert in an isolated AWS ML integration account, the following condition is observed: A model is deployed to a large general-purpose instance based only on notebook habits, with no comparison of lower-cost inference-optimized choices. The proposed change must preserve logging and least privilege. Which observation most directly identifies the root cause of the reported behavior?
- Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events.
- Measure memory, throughput, p95 latency and cost across supported candidate endpoint configurations.
- Inspect resource tags, cost allocation tag activation, Cost Explorer grouping, and unallocated resource types.
- Inspect group-level prediction and label distributions, subgroup sample sizes, and configured bias metrics.
Explanation
Correct Answer
B:Measure memory, throughput, p95 latency and cost across supported candidate endpoint configurations. This directly tests inference recommender benchmarking instance family in the reported AWS ML workflow. Rightsizing depends on workload measurements rather than the training machine or instance size alone. It distinguishes the problem before choosing a change. The related targeted response is: Use SageMaker Inference Recommender or equivalent controlled load tests to rightsize the serving instance family. Reported signal: A model is deployed to a large general-purpose instance based only on notebook habits, with no comparison of lower-cost inference-optimized choices. Operational setting: While reproducing the alert in an isolated AWS ML integration account.
Incorrect Answers
A:Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. This is a possible diagnostic for s3 sse-kms cross-account model artifact access, but it does not directly test the inference recommender benchmarking instance family failure reported here. Prioritize: Measure memory, throughput, p95 latency and cost across supported candidate endpoint configurations. The relevant case is: A model is deployed to a large general-purpose instance based only on notebook habits, with no comparison of lower-cost inference-optimized choices. Operational setting: While reproducing the alert in an isolated AWS ML integration account.
C:Inspect resource tags, cost allocation tag activation, Cost Explorer grouping, and unallocated resource types. This is a possible diagnostic for tag-based ml training cost attribution, but it does not directly test the inference recommender benchmarking instance family failure reported here. Prioritize: Measure memory, throughput, p95 latency and cost across supported candidate endpoint configurations. The relevant case is: A model is deployed to a large general-purpose instance based only on notebook habits, with no comparison of lower-cost inference-optimized choices. Operational setting: While reproducing the alert in an isolated AWS ML integration account.
D:Inspect group-level prediction and label distributions, subgroup sample sizes, and configured bias metrics. This is a possible diagnostic for clarify postdeployment group bias monitoring, but it does not directly test the inference recommender benchmarking instance family failure reported here. Prioritize: Measure memory, throughput, p95 latency and cost across supported candidate endpoint configurations. The relevant case is: A model is deployed to a large general-purpose instance based only on notebook habits, with no comparison of lower-cost inference-optimized choices. Operational setting: While reproducing the alert in an isolated AWS ML integration account.
Question 981
As a regulated ML workload prepares for an automated regional rollout, the following condition is observed: A regulated training process requires private network paths, but its isolated subnet cannot call the intended SageMaker API without going through public internet. The team must demonstrate a repeatable monitoring or security control. Which engineering response resolves the actual issue without masking the symptoms?
- Use a consistent versioned feature definition and test offline/online values at the same observation cutoff.
- Use stable user-cohort assignment or an appropriate experiment design while keeping exposed groups measurable.
- Use approved SageMaker interface endpoints and relevant private DNS and policy settings for required API access.
- Restrict writes to approved bucket prefixes and apply appropriate resource and organizational guardrails.
Explanation
Correct Answer
C:Use approved SageMaker interface endpoints and relevant private DNS and policy settings for required API access. This is the appropriate controlled remedy for private vpc endpoint policy for sagemaker apis. An IAM role permits API actions but network access and endpoint policies must also support the private path. Verify the cause with: Inspect subnet routes, VPC interface endpoints, private DNS settings, endpoint security groups, and endpoint policies. Then confirm: Training jobs reach the intended service privately while unauthorized external egress stays blocked. Reported signal: A regulated training process requires private network paths, but its isolated subnet cannot call the intended SageMaker API without going through public internet. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Incorrect Answers
A:Use a consistent versioned feature definition and test offline/online values at the same observation cutoff. This addresses a different issue involving training-serving feature skew incident. It does not correct the stated private vpc endpoint policy for sagemaker apis mechanism. The targeted response is: Use approved SageMaker interface endpoints and relevant private DNS and policy settings for required API access. Case evidence: A regulated training process requires private network paths, but its isolated subnet cannot call the intended SageMaker API without going through public internet. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
B:Use stable user-cohort assignment or an appropriate experiment design while keeping exposed groups measurable. This addresses a different issue involving champion challenger a/b assignment consistency. It does not correct the stated private vpc endpoint policy for sagemaker apis mechanism. The targeted response is: Use approved SageMaker interface endpoints and relevant private DNS and policy settings for required API access. Case evidence: A regulated training process requires private network paths, but its isolated subnet cannot call the intended SageMaker API without going through public internet. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
D:Restrict writes to approved bucket prefixes and apply appropriate resource and organizational guardrails. This addresses a different issue involving cross-account s3 bucket exfiltration protection. It does not correct the stated private vpc endpoint policy for sagemaker apis mechanism. The targeted response is: Use approved SageMaker interface endpoints and relevant private DNS and policy settings for required API access. Case evidence: A regulated training process requires private network paths, but its isolated subnet cannot call the intended SageMaker API without going through public internet. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Question 982
While reproducing a training incident in an independent SageMaker experiment, the following condition is observed: A team declares a new model better by comparing metrics from runs that used different target definitions and data partitions. The experiment must be compared fairly against its prior baseline. Which observation most directly identifies the root cause of the reported behavior?
- Profile model memory, throughput, communication overhead, and per-device utilization.
- Inspect accelerator memory use, tensor shapes, batch size, gradient accumulation, and failed training logs.
- Inspect Clarify attribution results, feature timestamp semantics, and whether the flag is created after the target event.
- Inspect experiment run lineage, dataset/version hashes, code revisions, random seeds, and metric definitions.
Explanation
Correct Answer
D:Inspect experiment run lineage, dataset/version hashes, code revisions, random seeds, and metric definitions. This directly tests reproducible sagemaker experiments comparisons in the reported AWS ML workflow. Performance claims are meaningful only when experiments control relevant changes and expose comparable evidence. It distinguishes the problem before choosing a change. The related targeted response is: Use tracked SageMaker Experiments runs with fixed comparable evaluation splits and a documented metric contract. Reported signal: A team declares a new model better by comparing metrics from runs that used different target definitions and data partitions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Incorrect Answers
A:Profile model memory, throughput, communication overhead, and per-device utilization. This is a possible diagnostic for data parallelism versus model parallelism, but it does not directly test the reproducible sagemaker experiments comparisons failure reported here. Prioritize: Inspect experiment run lineage, dataset/version hashes, code revisions, random seeds, and metric definitions. The relevant case is: A team declares a new model better by comparing metrics from runs that used different target definitions and data partitions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
B:Inspect accelerator memory use, tensor shapes, batch size, gradient accumulation, and failed training logs. This is a possible diagnostic for training batch size causes gpu memory exhaustion, but it does not directly test the reproducible sagemaker experiments comparisons failure reported here. Prioritize: Inspect experiment run lineage, dataset/version hashes, code revisions, random seeds, and metric definitions. The relevant case is: A team declares a new model better by comparing metrics from runs that used different target definitions and data partitions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
C:Inspect Clarify attribution results, feature timestamp semantics, and whether the flag is created after the target event. This is a possible diagnostic for clarify feature attribution against target leakage, but it does not directly test the reproducible sagemaker experiments comparisons failure reported here. Prioritize: Inspect experiment run lineage, dataset/version hashes, code revisions, random seeds, and metric definitions. The relevant case is: A team declares a new model better by comparing metrics from runs that used different target definitions and data partitions. Operational setting: While reproducing a training incident in an independent SageMaker experiment.
Question 983
During an operational load and release rehearsal for SageMaker-hosted inference, the following condition is observed: A platform maintains hundreds of low-traffic similar models and deploys each to a dedicated always-running instance, causing high idle cost. Change control requires measurable endpoint or pipeline acceptance evidence. Which investigation provides the most decisive FIRST evidence for the underlying problem?
- Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements.
- Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories.
- Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost.
- Compare template parameters, stack drift results, change sets, and deployed endpoint settings.
Explanation
Correct Answer
A:Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements. This directly tests multi-model endpoint hosting tradeoff in the reported AWS ML workflow. Multi-model hosting shares inference compute but can incur cache misses and is unsuitable when models require incompatible runtimes. It distinguishes the problem before choosing a change. The related targeted response is: Use a compatible SageMaker multi-model endpoint with managed model loading and monitor cache/eviction behavior. Reported signal: A platform maintains hundreds of low-traffic similar models and deploys each to a dedicated always-running instance, causing high idle cost. Operational setting: During an operational load and release rehearsal for SageMaker-hosted inference.
Incorrect Answers
B:Inspect EventBridge rule patterns, source events, object prefixes, and pipeline execution histories. This is a possible diagnostic for eventbridge conditional retraining trigger, but it does not directly test the multi-model endpoint hosting tradeoff failure reported here. Prioritize: Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements. The relevant case is: A platform maintains hundreds of low-traffic similar models and deploys each to a dedicated always-running instance, causing high idle cost. Operational setting: During an operational load and release rehearsal for SageMaker-hosted inference.
C:Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. This is a possible diagnostic for real-time versus batch inference decision, but it does not directly test the multi-model endpoint hosting tradeoff failure reported here. Prioritize: Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements. The relevant case is: A platform maintains hundreds of low-traffic similar models and deploys each to a dedicated always-running instance, causing high idle cost. Operational setting: During an operational load and release rehearsal for SageMaker-hosted inference.
D:Compare template parameters, stack drift results, change sets, and deployed endpoint settings. This is a possible diagnostic for cloudformation stack parameter drift, but it does not directly test the multi-model endpoint hosting tradeoff failure reported here. Prioritize: Inspect model sizes, framework compatibility, per-model request rates, load latency, and isolation requirements. The relevant case is: A platform maintains hundreds of low-traffic similar models and deploys each to a dedicated always-running instance, causing high idle cost. Operational setting: During an operational load and release rehearsal for SageMaker-hosted inference.
Question 984
During a controlled replay of the data-preparation workflow in a separate ML staging account, the following condition is observed: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. The dataset contract must remain reproducible for a backfill. Which engineering response resolves the actual issue without masking the symptoms?
- Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class.
- Write partitioned columnar Parquet data by event date, register accurate partitions in the catalog, and filter on partition keys.
- Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema.
- Repartition or salt the skewed join appropriately, or use a justified broadcast join when the small side fits memory.
Explanation
Correct Answer
C:Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. This is the appropriate controlled remedy for columnar file compaction for small objects. Excessive small files create metadata and request overhead even with a columnar data format. Verify the cause with: Measure file-size distribution, object request count, Spark task scheduling overhead, and read throughput. Then confirm: Training read time and object operations decline while the resulting row count and checksums match. Reported signal: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Incorrect Answers
A:Revise instructions, route ambiguous samples for expert adjudication, and monitor label quality per class. This addresses a different issue involving ground truth labeling agreement review. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
B:Write partitioned columnar Parquet data by event date, register accurate partitions in the catalog, and filter on partition keys. This addresses a different issue involving partition pruning in s3 parquet training input. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
D:Repartition or salt the skewed join appropriately, or use a justified broadcast join when the small side fits memory. This addresses a different issue involving glue spark join skew on dominant identifier. It does not correct the stated columnar file compaction for small objects mechanism. The targeted response is: Compact the partitioned dataset into appropriately sized Parquet objects without changing the record keys or schema. Case evidence: A daily S3 dataset consists of millions of tiny Parquet files, and training startup spends more time opening objects than scanning data. Operational setting: During a controlled replay of the data-preparation workflow in a separate ML staging account.
Question 985
While testing a second region for recoverable deployment of the same ML service, the following condition is observed: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. The deployment must be reproducible without manual console-only fixes. Select TWO: the appropriate diagnostic check and the targeted implementation change.
- Benchmark supported accelerated inference instances and choose GPU capacity only if latency and throughput gains justify cost.
- Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost.
- Grant the build role only required object and key operations for its specific artifact prefix and encryption key.
- Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow.
- Inspect the DAG of processing, training, evaluation and registration steps, along with parameter and artifact lineage.
Explanation
Correct Answers
B:Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. This directly tests real-time versus batch inference decision in the reported AWS ML workflow. Batch inference suits asynchronous bulk datasets without an interactive low-latency response requirement. It distinguishes the problem before choosing a change. The related targeted response is: Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. Operational setting: While testing a second region for recoverable deployment of the same ML service.
D:Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow. This is the appropriate controlled remedy for real-time versus batch inference decision. Batch inference suits asynchronous bulk datasets without an interactive low-latency response requirement. Verify the cause with: Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. Then confirm: The overnight batch completes within its service window while there is no idle always-on inference endpoint charge. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Incorrect Answers
A:Benchmark supported accelerated inference instances and choose GPU capacity only if latency and throughput gains justify cost. This selection concerns gpu versus cpu inference fit rather than the required diagnostic-and-remedy pair for real-time versus batch inference decision. Use evidence: Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. and the matching correction: Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow. Signal: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. Operational setting: While testing a second region for recoverable deployment of the same ML service.
C:Grant the build role only required object and key operations for its specific artifact prefix and encryption key. This selection concerns codebuild iam least-privilege artifact upload rather than the required diagnostic-and-remedy pair for real-time versus batch inference decision. Use evidence: Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. and the matching correction: Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow. Signal: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. Operational setting: While testing a second region for recoverable deployment of the same ML service.
E:Inspect the DAG of processing, training, evaluation and registration steps, along with parameter and artifact lineage. This selection concerns sagemaker pipelines parameterized retraining rather than the required diagnostic-and-remedy pair for real-time versus batch inference decision. Use evidence: Inspect invocation cadence, acceptable completion window, per-record latency requirement, and idle endpoint cost. and the matching correction: Use a SageMaker Batch Transform or batch inference job for the scheduled bulk scoring workflow. Signal: A customer scoring workload runs only once nightly over millions of S3 records, but a team provisions a permanently running low-latency endpoint for it. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Question 986
During a production MLOps reliability and security review, the following condition is observed: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. The owner needs an auditable resolution without broadening privileges. What targeted change should be implemented after verifying the stated cause?
- Set budget alerts and bounded tuning job quotas before launching experiments, and enforce an owner-approved stopping policy.
- Pin an immutable approved artifact version or content digest and verify integrity before deployment.
- Identify the component contributing to total request time and optimize network or overhead separately from actual model inference.
- Use appropriate SageMaker Clarify model-bias monitoring and investigate supported group differences before deployment changes.
Explanation
Correct Answer
C:Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. This is the appropriate controlled remedy for cloudwatch invocations and modellatency distinction. Model compute time and overall endpoint latency are distinct signals; blaming model code without tracing overhead is premature. Verify the cause with: Compare ModelLatency with OverheadLatency, client network timing, invocation counts, and endpoint 5XX responses. Then confirm: The chosen fix reduces measured end-to-end p95 latency and component metrics explain the change. Reported signal: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Incorrect Answers
A:Set budget alerts and bounded tuning job quotas before launching experiments, and enforce an owner-approved stopping policy. This addresses a different issue involving aws budgets spend threshold for experimentation. It does not correct the stated cloudwatch invocations and modellatency distinction mechanism. The targeted response is: Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. Case evidence: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
B:Pin an immutable approved artifact version or content digest and verify integrity before deployment. This addresses a different issue involving model artifact supply-chain integrity. It does not correct the stated cloudwatch invocations and modellatency distinction mechanism. The targeted response is: Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. Case evidence: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
D:Use appropriate SageMaker Clarify model-bias monitoring and investigate supported group differences before deployment changes. This addresses a different issue involving clarify postdeployment group bias monitoring. It does not correct the stated cloudwatch invocations and modellatency distinction mechanism. The targeted response is: Identify the component contributing to total request time and optimize network or overhead separately from actual model inference. Case evidence: An endpoint's client-side response time rises, but the team only examines SageMaker ModelLatency and concludes the model is slow. Operational setting: During a production MLOps reliability and security review.
Question 987
During a production MLOps reliability and security review, the following condition is observed: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. The owner needs an auditable resolution without broadening privileges. Which TWO complementary choices provide the decisive diagnosis AND corrective response? Select TWO.
- Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics.
- Use SageMaker Inference Recommender or equivalent controlled load tests to rightsize the serving instance family.
- Restrict writes to approved bucket prefixes and apply appropriate resource and organizational guardrails.
- Inspect online versus offline feature computation code, event-time cutoffs, freshness, and sampled entity values.
- Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age.
Explanation
Correct Answers
A:Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. This directly tests asynchronous inference queue backlog alert in the reported AWS ML workflow. Queued work can miss response deadlines without saturating compute; queue age is its own performance signal. It distinguishes the problem before choosing a change. The related targeted response is: Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: During a production MLOps reliability and security review.
E:Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. This is the appropriate controlled remedy for asynchronous inference queue backlog alert. Queued work can miss response deadlines without saturating compute; queue age is its own performance signal. Verify the cause with: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. Then confirm: Backlog age stabilizes and requests complete within the agreed service-level window. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: During a production MLOps reliability and security review.
Incorrect Answers
B:Use SageMaker Inference Recommender or equivalent controlled load tests to rightsize the serving instance family. This selection concerns inference recommender benchmarking instance family rather than the required diagnostic-and-remedy pair for asynchronous inference queue backlog alert. Use evidence: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. and the matching correction: Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. Signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: During a production MLOps reliability and security review.
C:Restrict writes to approved bucket prefixes and apply appropriate resource and organizational guardrails. This selection concerns cross-account s3 bucket exfiltration protection rather than the required diagnostic-and-remedy pair for asynchronous inference queue backlog alert. Use evidence: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. and the matching correction: Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. Signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: During a production MLOps reliability and security review.
D:Inspect online versus offline feature computation code, event-time cutoffs, freshness, and sampled entity values. This selection concerns training-serving feature skew incident rather than the required diagnostic-and-remedy pair for asynchronous inference queue backlog alert. Use evidence: Inspect async inference backlog size, age of oldest request, worker capacity and completion/error metrics. and the matching correction: Scale compatible asynchronous inference capacity or tune workload and queue alarms for pending-request age. Signal: An asynchronous inference queue's oldest pending request age keeps increasing, yet GPU utilization alerts remain normal because requests are waiting before execution. Operational setting: During a production MLOps reliability and security review.
Question 988
At a production AWS ML data lake supporting customer-facing predictions, the following condition is observed: A raw CSV field contains quoted commas and a header is sometimes missing, so naïve delimiter splitting shifts the label into the wrong column. Engineers must avoid changing unrelated source partitions. Which observation most directly identifies the root cause of the reported behavior?
- Inspect feature availability timestamps and whether each label time follows the last feature observation.
- Inspect preprocessing fit order, imputation statistics, dataset split boundaries, and pipeline serialization.
- Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint.
- Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job.
Explanation
Correct Answer
D:Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job. This directly tests csv schema ambiguity during ingestion in the reported AWS ML workflow. CSV parsing must respect quoting and schema; splitting on every comma corrupts valid quoted fields. It distinguishes the problem before choosing a change. The related targeted response is: Use a CSV parser with explicit quoting and schema validation; quarantine malformed records before training. Reported signal: A raw CSV field contains quoted commas and a header is sometimes missing, so naïve delimiter splitting shifts the label into the wrong column. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Incorrect Answers
A:Inspect feature availability timestamps and whether each label time follows the last feature observation. This is a possible diagnostic for chronological split for forecast evaluation, but it does not directly test the csv schema ambiguity during ingestion failure reported here. Prioritize: Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job. The relevant case is: A raw CSV field contains quoted commas and a header is sometimes missing, so naïve delimiter splitting shifts the label into the wrong column. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
B:Inspect preprocessing fit order, imputation statistics, dataset split boundaries, and pipeline serialization. This is a possible diagnostic for median imputation fitted on training split, but it does not directly test the csv schema ambiguity during ingestion failure reported here. Prioritize: Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job. The relevant case is: A raw CSV field contains quoted commas and a header is sometimes missing, so naïve delimiter splitting shifts the label into the wrong column. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
C:Check job names, bookmark settings, run history, source object timestamps, and the previously committed checkpoint. This is a possible diagnostic for glue job bookmark state error, but it does not directly test the csv schema ambiguity during ingestion failure reported here. Prioritize: Inspect sample raw records, quoting rules, expected headers, and parsed column counts in the ingestion job. The relevant case is: A raw CSV field contains quoted commas and a header is sometimes missing, so naïve delimiter splitting shifts the label into the wrong column. Operational setting: At a production AWS ML data lake supporting customer-facing predictions.
Question 989
While testing a second region for recoverable deployment of the same ML service, the following condition is observed: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. The deployment must be reproducible without manual console-only fixes. Which TWO steps best establish the evidence and correct the identified root cause? Select TWO.
- Require model evaluation and explicit approval gates before production deployment, blocking failed metric thresholds.
- Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory.
- Inspect the DAG of processing, training, evaluation and registration steps, along with parameter and artifact lineage.
- Implement the SageMaker inference container
/ping and /invocations expectations or use a supported serving toolkit. - Reference an approved immutable image digest or controlled version tag and retain a reproducible build record.
Explanation
Correct Answers
B:Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory. This directly tests ecr immutable image digest for inference in the reported AWS ML workflow. Mutable container tags can point to different images; reproducible ML deployment needs immutable artifact references. It distinguishes the problem before choosing a change. The related targeted response is: Reference an approved immutable image digest or controlled version tag and retain a reproducible build record. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. Operational setting: While testing a second region for recoverable deployment of the same ML service.
E:Reference an approved immutable image digest or controlled version tag and retain a reproducible build record. This is the appropriate controlled remedy for ecr immutable image digest for inference. Mutable container tags can point to different images; reproducible ML deployment needs immutable artifact references. Verify the cause with: Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory. Then confirm: Identical deployment revisions resolve to the same container image digest and environment dependencies. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Incorrect Answers
A:Require model evaluation and explicit approval gates before production deployment, blocking failed metric thresholds. This selection concerns codepipeline approval gate on model metrics rather than the required diagnostic-and-remedy pair for ecr immutable image digest for inference. Use evidence: Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory. and the matching correction: Reference an approved immutable image digest or controlled version tag and retain a reproducible build record. Signal: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. Operational setting: While testing a second region for recoverable deployment of the same ML service.
C:Inspect the DAG of processing, training, evaluation and registration steps, along with parameter and artifact lineage. This selection concerns sagemaker pipelines parameterized retraining rather than the required diagnostic-and-remedy pair for ecr immutable image digest for inference. Use evidence: Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory. and the matching correction: Reference an approved immutable image digest or controlled version tag and retain a reproducible build record. Signal: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. Operational setting: While testing a second region for recoverable deployment of the same ML service.
D:Implement the SageMaker inference container /ping and /invocations expectations or use a supported serving toolkit. This selection concerns bring-your-own-inference-container contract rather than the required diagnostic-and-remedy pair for ecr immutable image digest for inference. Use evidence: Inspect ECR image digest, tag history, pipeline manifests, and built-image package inventory. and the matching correction: Reference an approved immutable image digest or controlled version tag and retain a reproducible build record. Signal: A model endpoint repeatedly deploys an image tagged latest, causing unexpectedly different runtime libraries after each redeploy. Operational setting: While testing a second region for recoverable deployment of the same ML service.
Question 990
As a regulated ML workload prepares for an automated regional rollout, the following condition is observed: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. The team must demonstrate a repeatable monitoring or security control. Which TWO steps best establish the evidence and correct the identified root cause? Select TWO.
- Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events.
- Use structured logging and scoped CloudWatch Logs Insights queries to aggregate failures by training job and time.
- Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access.
- Enable appropriate CloudTrail S3 data events for the sensitive dataset with protected storage and retention.
- Inspect captured request IDs, ground-truth keys, and the quality monitor dataset schema.
Explanation
Correct Answers
A:Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. This directly tests s3 sse-kms cross-account model artifact access in the reported AWS ML workflow. S3 object permission does not imply access to the KMS encryption key used on the object. It distinguishes the problem before choosing a change. The related targeted response is: Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
C:Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access. This is the appropriate controlled remedy for s3 sse-kms cross-account model artifact access. S3 object permission does not imply access to the KMS encryption key used on the object. Verify the cause with: Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. Then confirm: The endpoint loads the intended encrypted artifact while unrelated principals cannot decrypt it. This is one part of a select-two decision; the matching evidence and corrective action are both required. Reported signal: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
Incorrect Answers
B:Use structured logging and scoped CloudWatch Logs Insights queries to aggregate failures by training job and time. This selection concerns cloudwatch logs insights error correlation rather than the required diagnostic-and-remedy pair for s3 sse-kms cross-account model artifact access. Use evidence: Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. and the matching correction: Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access. Signal: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
D:Enable appropriate CloudTrail S3 data events for the sensitive dataset with protected storage and retention. This selection concerns cloudtrail training data access audit coverage rather than the required diagnostic-and-remedy pair for s3 sse-kms cross-account model artifact access. Use evidence: Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. and the matching correction: Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access. Signal: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
E:Inspect captured request IDs, ground-truth keys, and the quality monitor dataset schema. This selection concerns model monitor ground truth association rather than the required diagnostic-and-remedy pair for s3 sse-kms cross-account model artifact access. Use evidence: Inspect S3 object encryption, key policy, grant permissions, IAM role, and denied KMS decrypt events. and the matching correction: Authorize the specific serving role to decrypt with the relevant KMS key under a restricted key policy and S3 access. Signal: A SageMaker endpoint role has s3:GetObject for a model artifact, but download fails because the artifact is encrypted under a KMS key in a different account. Operational setting: As a regulated ML workload prepares for an automated regional rollout.
