Question 190
A new supplier sends literal 'unknown' in an amount field and the ETL cast now fails. Which correction is appropriate?
- Executor memory tuning
- Spark shuffle skew investigation
- Data type casting validation
- Glue error log inspection
Explanation
Correct Answer
C:Explicit type validation and casting prevent malformed input strings from corrupting typed transformations; in this scenario, validate or quarantine malformed values using a documented rule, rather than silently coercing every row to an arbitrary default.
Incorrect Answers
A:Spark executor memory and partition size affect whether large tasks spill or fail from memory pressure; its benefit is addressing executor out-of-memory errors on large partitions; that is different from handling unexpected nonnumeric values in a numeric column.
B:Spark stage metrics can reveal straggler tasks and imbalanced shuffle partitions in a join; the current need is for handling unexpected nonnumeric values in a numeric column, not finding a slow join caused by one hot key.
D:Glue job failures produce execution logs that distinguish syntax, credentials, schema and runtime failures; this helps with locating the precise failing operation in Glue, not handling unexpected nonnumeric values in a numeric column.
Question 191
Analysts need a project workspace to inspect governed datasets and collaborate on preparation before publishing curated tables. Which experience fits?
- SageMaker Unified Studio data exploration
- Glue DataBrew recipe
- DataBrew profiling job
- Lambda event-driven preprocessing
- Glue Spark preprocessing job
Explanation
Correct Answer
A:Unified Studio provides a governed collaborative environment for exploring and preparing datasets; in this scenario, use authorized project assets and data tools within Unified Studio rather than copying raw data into unmanaged personal workspaces.
Incorrect Answers
B:DataBrew recipes apply documented preparation steps such as normalizing and cleaning tabular fields; although useful for letting analysts prepare messy CSV columns without writing Spark code, it leaves the need for preparing datasets in a shared analytics project unmet.
C:A DataBrew profile job measures dataset properties including value patterns, completeness and distributions; it would target inspecting quality problems before deciding cleaning rules, while the question is about preparing datasets in a shared analytics project.
D:Lambda can perform small validation or normalization tasks when an object-arrival event is received; its benefit is normalizing small files immediately after landing; that is different from preparing datasets in a shared analytics project.
E:Glue Spark jobs transform larger datasets with joins, repartitioning and schema mapping; the current need is for preparing datasets in a shared analytics project, not preparing large multi-file datasets for analytics.
Question 192
A single host cannot process a multiterabyte event archive within the required batch window. What design principle helps?
- Lambda Amazon EFS mount
- Graph traversal algorithm
- Distributed partitioned compute
- Lambda /tmp ephemeral storage
Explanation
Correct Answer
C:Distributed processing partitions data so workers handle portions in parallel and combine results as needed; in this scenario, partition the input for parallel execution and manage shuffle and coordination costs when merging partial results.
Incorrect Answers
A:Lambda can mount an EFS file system through access points from appropriately networked functions; although useful for sharing persistent files among multiple Lambda executions, it leaves the need for parallelizing a large aggregation beyond one machine unmet.
B:Graph traversal explores connected vertices and edges to follow relationships or dependencies; it would target following lineage dependencies between datasets, while the question is about parallelizing a large aggregation beyond one machine.
D:Lambda provides configurable temporary /tmp storage local to an execution environment; the current need is for parallelizing a large aggregation beyond one machine, not buffering intermediate files during one function execution.
Question 193
An acquired platform already uses Kafka topics, consumer groups, and Kafka APIs; operators want to avoid running broker infrastructure. What should host ingestion?
- Amazon Data Firehose
- Amazon Kinesis Data Streams
- Amazon MSK
- DynamoDB Streams
Explanation
Correct Answer
C:Amazon MSK runs managed Apache Kafka brokers while retaining Kafka client and topic semantics; in this scenario, mSK preserves Kafka compatibility and delegates broker operations to AWS while existing producers and readers retain their integration pattern.
Incorrect Answers
A:Firehose buffers streaming records and delivers them to configured analytics destinations; although useful for continuously depositing events into an S3 data lake, it leaves the need for migrating existing Kafka producers without rewriting clients unmet.
B:A data stream retains ordered records in shards for applications that read at their own pace; it would target keeping a replayable clickstream for independent consumers, while the question is about migrating existing Kafka producers without rewriting clients.
D:DynamoDB Streams exposes item-level changes made to a DynamoDB table for downstream processors; the current need is for migrating existing Kafka producers without rewriting clients, not reacting to inserts and updates of table items.
Question 194
A research lake must retain SQL exports, PDFs and nested event documents before curated processing. Which V is emphasized?
- High-velocity streaming architecture
- High-volume batch architecture
- Data veracity validation
- Schema-flexible variety handling
Explanation
Correct Answer
D:Variety covers multiple structures and types including relational rows, JSON, images and free text; in this scenario, plan for heterogeneous formats and metadata; enforce schemas only where the downstream processing requires them.
Incorrect Answers
A:Velocity describes how rapidly data arrives and how quickly it needs to be handled; its benefit is coping with millions of events arriving per second; that is different from ingesting incompatible structured and unstructured formats.
B:Volume describes the total quantity of data that must be stored and processed; the current need is for ingesting incompatible structured and unstructured formats, not scaling storage and parallelism for terabytes of historical data.
C:Veracity concerns trustworthiness and reliability of incoming information rather than its quantity or speed; this can support checking suspicious and conflicting source records, but would not address the need for ingesting incompatible structured and unstructured formats.
Question 195
An analyst wants to correlate long-lived S3 trail archives with business reference tables using SQL. Which analytics engine fits?
- CloudWatch Logs subscription forwarding
- Athena over S3 CloudTrail logs
- CloudTrail Lake SQL query
- OpenSearch Service indexed audit exploration
- CloudWatch Logs Insights query
Explanation
Correct Answer
B:Athena can query CloudTrail files in S3 after defining compatible schemas and partitions; in this scenario, catalog the raw log format and query through Athena, partitioning and restricting access appropriately.
Incorrect Answers
A:Subscription filters can deliver matching CloudWatch log events to downstream processing destinations; it would target streaming application audit records to a central pipeline, while the question is about joining archived API logs with lake reference datasets.
C:CloudTrail Lake supports SQL-based analysis over events collected in event data stores; the current need is for joining archived API logs with lake reference datasets, not investigating account activity with structured SQL.
D:OpenSearch can index log documents to support interactive search and visualization; this can support searching repeated substrings across a large normalized log index, but would not address the need for joining archived API logs with lake reference datasets.
E:Logs Insights performs interactive filtered searches and aggregations over CloudWatch application log groups; this helps with counting application exception types in Lambda logs, not joining archived API logs with lake reference datasets.
Question 196
Compliance needs assurance that CloudTrail files in S3 can be checked for integrity after collection. What feature should be enabled?
- CloudTrail S3 object data events
- CloudTrail log file integrity validation
- CloudTrail management-event records
- CloudTrail organization trail
Explanation
Correct Answer
B:Trail log file integrity validation provides digest files that can help detect tampering of delivered log files; in this scenario, enable delivery of digest files and perform validation with trusted tooling and controlled storage access.
Incorrect Answers
A:CloudTrail can capture selected S3 object-level APIs as data events when explicitly configured; it would target identifying who downloaded restricted data-lake objects, while the question is about proving stored audit logs were not silently altered.
C:Management events track AWS control-plane actions such as PutBucketPolicy and CreateRole; the current need is for proving stored audit logs were not silently altered, not finding who changed resource permissions.
D:An organization trail can collect supported management activity centrally for accounts in an AWS Organization; this can support covering many accounts with one audit policy, but would not address the need for proving stored audit logs were not silently altered.
Question 197
A Glue crawl and two ETL steps must use the same daily batch ID without hard-coded dates in each job. Which native mechanism helps?
- Airflow catchup configuration
- Glue workflow run properties
- Step Functions task token callback
- EventBridge Scheduler retry policy
Explanation
Correct Answer
B:Glue workflow run properties allow pipeline components to share run-scoped configuration values; in this scenario, set run properties at workflow start and read them in participating jobs for a consistent batch-scoped reference.
Incorrect Answers
A:Airflow catchup behavior determines whether a DAG schedules missed historical intervals when enabled; it would target preventing a flood of old scheduled DAG runs, while the question is about passing an ingestion batch identifier between Glue jobs.
C:A task token lets a state-machine task wait for an external system to signal completion; the current need is for passing an ingestion batch identifier between Glue jobs, not pausing a workflow for asynchronous external approval.
D:Scheduler targets can have retry policies and dead-letter queue configuration for failed delivery attempts; this can support recovering from temporary failure to invoke a scheduled target, but would not address the need for passing an ingestion batch identifier between Glue jobs.
Question 198
A remote observatory has petabytes of historical captures but only a slow and unreliable WAN link. Which migration approach can reduce transfer time?
- AWS Transfer Family SFTP server
- AWS Snow Family offline transfer
- AWS DataSync transfer task
- S3 multipart upload
Explanation
Correct Answer
B:Snow devices can support offline movement of large datasets when network bandwidth is inadequate; in this scenario, stage data using an appropriate Snow service workflow and import it without depending solely on the constrained network.
Incorrect Answers
A:Transfer Family exposes managed SFTP endpoints that can land partner files into S3 or EFS; this can support receiving external partner uploads over SFTP, but would not address the need for seeding a cloud lake from a bandwidth-constrained site.
C:DataSync moves large datasets between supported on-premises or AWS storage locations with transfer management; although useful for migrating terabytes from an on-premises file share into S3, it leaves the need for seeding a cloud lake from a bandwidth-constrained site unmet.
D:Multipart upload sends large objects as separately uploaded parts that can be retried independently; it would target reliably uploading individual multi-gigabyte files over a variable connection, while the question is about seeding a cloud lake from a bandwidth-constrained site.
Question 199
A team needs to associate a preprocessing artifact with the dataset and processing execution used in an ML pipeline. Which lineage tool is relevant?
- Job-run lineage metadata
- SageMaker Catalog lineage graph
- Column-level lineage mapping
- SageMaker ML Lineage Tracking
Explanation
Correct Answer
D:ML lineage facilities can record relationships among model-development artifacts and processing steps; in this scenario, record supported lineage associations between sources, processing activities and artifacts rather than relying on filenames alone.
Incorrect Answers
A:Recording run identifiers, source versions and target outputs connects a materialized dataset to its producing workflow; this helps with auditing the exact files used in one derived table version, not tracing artifacts in a supported ML-related data preparation workflow.
B:Catalog lineage can represent data assets and transformations to help trace upstream and downstream dependencies; although useful for tracing business report fields to source datasets, it leaves the need for tracing artifacts in a supported ML-related data preparation workflow unmet.
C:Column-level lineage records how source fields produce downstream columns through transformations; it would target determining whether a sensitive input field reaches a report column, while the question is about tracing artifacts in a supported ML-related data preparation workflow.
Question 200
A state machine invokes a throttled service that usually recovers within seconds. Which error-handling design is appropriate?
- SQS ChangeMessageVisibility heartbeat
- Idempotent processing checkpoint
- SQS dead-letter queue
- Step Functions retry with backoff
- Multi-AZ dependent services
Explanation
Correct Answer
D:State machine Retry rules can handle selected errors with intervals and exponential backoff; in this scenario, define a selective Retry policy and a failure path so short-lived errors are retried without restarting the entire pipeline.
Incorrect Answers
A:The SQS ChangeMessageVisibility API can dynamically extend the invisibility period of an in-flight message as a variable-length task proceeds; although useful for extending SQS visibility for a task whose running time varies widely, it leaves the need for recovering from transient downstream API failures unmet.
B:Durable checkpoints plus idempotent sink writes limit duplicate effects after task retries; it would target making a restarted pipeline safe to replay, while the question is about recovering from transient downstream API failures.
C:A dead-letter queue collects messages that repeatedly fail processing after configured receive attempts; its benefit is isolating permanently failing messages for investigation; that is different from recovering from transient downstream API failures.
E:Using supported multi-AZ data services can reduce interruption from a single Availability Zone failure; this can support keeping stateful dependencies available during one AZ outage, but would not address the need for recovering from transient downstream API failures.