Question 70

Marketing IDs and payment IDs use different numbering schemes, producing mismatched records in a join. What design step is needed?

  1. Deduplication across sources
  2. Late-arriving dimension handling
  3. Time zone normalization
  4. Stable join key

Explanation

Correct Answer

D:

Joining data from separate systems requires a trustworthy key that identifies the same entity on each side; in this scenario, create or use an authoritative crosswalk key, validate uniqueness and normalize source identifiers before joining.

Incorrect Answers

A:

Entity deduplication uses an agreed business key and conflict-resolution rule across integrated sources; this can support merging duplicate customer updates from two systems, but would not address the need for preventing incorrect joins across customer feeds.

B:

Late-arriving dimensions require a defined strategy for facts received before their lookup rows; this helps with loading facts that precede their dimension records, not preventing incorrect joins across customer feeds.

C:

Normalizing timestamps to a common time basis prevents false ordering across systems using different zones; although useful for correlating event timestamps from multiple geographic sources, it leaves the need for preventing incorrect joins across customer feeds unmet.

Question 71

An Airflow extract sometimes fails transiently and must retry automatically; a cleanup task must still run after the final retry fails. Which TWO Airflow task configurations meet both requirements? Choose TWO.

  1. Step Functions execution history
  2. Glue conditional trigger
  3. EventBridge Scheduler flexible time window
  4. Airflow task retry policy
  5. Airflow dependency trigger rule

Explanation

Correct Answers

D:

Airflow tasks support retry counts and delays within a DAG's execution schedule; in this scenario, configure task retries with appropriate delay while keeping writes safe when the same task is attempted again.

E:

Airflow trigger rules determine when a downstream task may run relative to upstream task states; for the additional requirement of ensuring a cleanup task runs after upstream failures, set an appropriate downstream trigger rule rather than the default all-success dependency behavior.

Incorrect Answers

A:

Step Functions maintains execution event history for supported workflows and helps locate failed states; although useful for tracing the precise failure transition in a workflow, it leaves the need for retrying a transiently failing DAG task unmet.

B:

Glue conditional triggers can start downstream jobs when designated upstream jobs or crawlers reach configured states; it would target running a publish job only after upstream Glue success, while the question is about retrying a transiently failing DAG task.

C:

Flexible time windows let a scheduled invocation occur within a configured interval instead of at an exact instant; its benefit is spreading noncritical batch launches to avoid thundering herds; that is different from retrying a transiently failing DAG task.

Question 72

A pipeline-failure event must be sent to multiple subscribers and leave a durable work item for a remediation worker that may be offline. Which TWO messaging services meet both requirements? Choose TWO.

  1. SNS topic fan-out
  2. Step Functions workflow status alert
  3. SQS durable queue
  4. CloudWatch alarm action
  5. EventBridge event-to-target routing

Explanation

Correct Answers

A:

SNS publishes notifications to multiple subscribed endpoints or queues using topic subscriptions; in this scenario, publish once to an SNS topic and configure appropriate subscriptions, keeping recipients independent of the source workflow.

C:

SQS retains messages for asynchronous processing by consumers that may be temporarily unavailable; for the additional requirement of buffering remediation work until a worker can run, queue a durable work item and process it with acknowledgments and retries rather than assuming a live subscriber.

Incorrect Answers

B:

An EventBridge rule can match Step Functions execution status changes for monitoring and notifications; it would target notifying when a whole workflow fails rather than one retrying task, while the question is about delivering one alert to several independent recipients.

D:

CloudWatch alarms evaluate metrics against thresholds and can invoke configured notification actions; the current need is for delivering one alert to several independent recipients, not alerting after repeated error metric breaches.

E:

EventBridge rules route matching state-change events to designated consumers; this can support notifying only after a specific job state transition, but would not address the need for delivering one alert to several independent recipients.

Question 73

A quality audit must guarantee meaningful sample coverage of rare small-country records as well as a dominant region. Which sampling method fits best?

  1. Stratified random sampling
  2. Random sampling
  3. Reservoir sampling
  4. Spark key salting
  5. Systematic sampling caveat

Explanation

Correct Answer

A:

Stratified sampling draws observations from defined subgroups to preserve rare but important categories; in this scenario, sample independently within defined country strata so rare populations are represented at the needed confidence.

Incorrect Answers

B:

Simple random sampling selects records with equal inclusion probability when an unbiased population estimate is needed; this helps with building a general quality sample without subgroup constraints, not assessing a dataset with very uneven country distributions.

C:

Reservoir sampling can maintain a bounded random sample from a stream of unknown total length; although useful for sampling a large unknown-length event stream with fixed memory, it leaves the need for assessing a dataset with very uneven country distributions unmet.

D:

Key salting distributes records for a heavily skewed join key across multiple synthetic partitions; it would target reducing one-key hotspots in a distributed join, while the question is about assessing a dataset with very uneven country distributions.

E:

Systematic sampling takes observations at regular intervals and may bias results when the stream is periodic; its benefit is avoiding periodic bias when checking machine readings; that is different from assessing a dataset with very uneven country distributions.

Question 74

A daily query selects all fields from a wide Parquet dataset though only three columns affect the result. What change reduces scan cost?

  1. Column projection before scan
  2. Glue worker sizing and job profiling
  3. Partition pruning
  4. EMR managed scaling

Explanation

Correct Answer

A:

Reading only required fields can reduce bytes processed by columnar storage and SQL queries; in this scenario, project the needed columns early so engines can avoid unnecessary column data and reduce I/O.

Incorrect Answers

B:

Glue worker types and counts affect job parallelism, memory and billed compute consumption; this can support right-sizing an oversized Glue ETL run, but would not address the need for minimizing scanned data during analytic transformations.

C:

Partition pruning excludes dataset partitions whose values cannot satisfy a query predicate; this helps with avoiding reads of unrelated months in a data lake, not minimizing scanned data during analytic transformations.

D:

EMR managed scaling can adjust cluster capacity within configured limits based on workload demand; although useful for reducing idle cluster capacity between Spark stages, it leaves the need for minimizing scanned data during analytic transformations unmet.

Question 75

An Athena fleet and multiple Glue jobs need one AWS-native metadata source for S3 tables. Which catalog is appropriate?

  1. AWS Glue Data Catalog
  2. Catalog API lookup
  3. Apache Hive metastore
  4. Catalog permissions boundary

Explanation

Correct Answer

A:

Glue Data Catalog is a managed technical metadata repository for datasets and ETL integration; in this scenario, register databases, tables, schemas and partitions there while using separate access controls for underlying data.

Incorrect Answers

B:

A data-processing application can query catalog APIs to discover schemas and data locations at runtime; its benefit is avoiding hard-coded column lists in a dynamic ETL client; that is different from centralizing technical table and partition definitions.

C:

Hive metastore maintains table and partition metadata for compatible Hadoop ecosystem tools; the current need is for centralizing technical table and partition definitions, not retaining a legacy Hive-compatible metadata system.

D:

Access to technical metadata and access to stored objects may be separately controlled by IAM and Lake Formation; this can support diagnosing why a visible table cannot be read, but would not address the need for centralizing technical table and partition definitions.

Question 76

A date-partitioned Athena external table has many missing catalog partitions, although S3 follows Hive-style key=value directories. Which SQL maintenance command can discover them automatically?

  1. ALTER TABLE ADD PARTITION
  2. Glue crawler partition update
  3. MSCK REPAIR TABLE
  4. Partition pruning predicate
  5. Athena partition projection

Explanation

Correct Answer

C:

Athena's MSCK REPAIR TABLE can discover compatible Hive-style S3 partitions and add missing catalog entries; in this scenario, run partition repair where supported to discover partition directories and register their metadata; consider scale for very large partition counts.

Incorrect Answers

A:

Adding a partition explicitly registers specified partition values and their locations in the catalog; this can support registering a small known list of partitions precisely, but would not address the need for registering newly created Hive-style partition folders.

B:

A Glue crawler can discover supported new partitions and maintain catalog metadata during a run; this helps with keeping partition metadata synchronized after scheduled ingestion, not registering newly created Hive-style partition folders.

D:

Query filters on partition columns allow compatible engines to avoid scanning unrelated object prefixes; it would target reducing reads after partition registration, while the question is about registering newly created Hive-style partition folders.

E:

Partition projection computes partition locations from table properties instead of storing every partition explicitly; its benefit is avoiding per-day catalog updates for predictable partitions; that is different from registering newly created Hive-style partition folders.

Question 77

A Step Functions pipeline must process thousands of manifest entries in parallel, branch between XML and JSON handlers, and quarantine failures after retries. Which THREE state features support these independent requirements? Choose THREE.

  1. Step Functions execution input/output paths
  2. Lambda timeout setting
  3. Step Functions Catch transition
  4. Step Functions Choice state
  5. Step Functions Map state

Explanation

Correct Answers

C:

Catch handlers move execution to a recovery state when a configured task error occurs; for the additional requirement of sending failed processing to a quarantine path, define a Catch transition to a quarantine or notification step, retaining error detail for later review.

D:

A Choice state routes execution based on a condition in the state input; for the additional requirement of selecting a branch after checking data type, evaluate the detected format and dispatch to the matching branch without creating an external polling controller.

E:

A Map state processes collection items using controlled parallel iterations in a workflow; in this scenario, use a Map iteration mode with configured concurrency and error handling to process independent objects under one execution.

Incorrect Answers

A:

InputPath, ResultPath and OutputPath can shape data passed between workflow states; this can support retaining only relevant fields between processing tasks, but would not address the need for applying the same transform to many independent input files.

B:

A Lambda timeout caps the maximum duration of a single function invocation; this helps with preventing a stalled serverless task from running indefinitely, not applying the same transform to many independent input files.

Question 78

An incident responder needs an ad hoc interactive query that groups the last twenty minutes of Lambda log events by error type and lets them inspect matching log lines without first exporting records to S3. Which tool fits?

  1. OpenSearch indexed log search
  2. Athena over structured S3 logs
  3. CloudWatch subscription filter
  4. CloudWatch metric filter
  5. CloudWatch Logs Insights

Explanation

Correct Answer

E:

Logs Insights searches and aggregates records in CloudWatch log groups with query syntax; in this scenario, use CloudWatch Logs Insights against the Lambda log groups with parsed fields, filter expressions and stats aggregation to investigate the errors.

Incorrect Answers

A:

OpenSearch can index log documents for search, filtering and operational dashboards; although useful for searching high-cardinality text patterns across indexed application logs, it leaves the need for interactively grouping recent Lambda exception events from CloudWatch log groups unmet.

B:

Athena can query centrally stored S3 audit logs after their schema is represented in a catalog; it would target running SQL over accumulated S3 log archives, while the question is about interactively grouping recent Lambda exception events from CloudWatch log groups.

C:

Log subscription filters send matching events from CloudWatch Logs to supported downstream consumers; its benefit is streaming selected application errors into a central processor; that is different from interactively grouping recent Lambda exception events from CloudWatch log groups.

D:

A metric filter creates CloudWatch metric datapoints from matching CloudWatch Logs events; the current need is for interactively grouping recent Lambda exception events from CloudWatch log groups, not alerting on a distinctive application error pattern.

Question 79

A Glue transform succeeds technically but sometimes rejects thousands of malformed records; operators want an alarm on rejection count. What telemetry fits?

  1. CloudWatch Logs retention policy
  2. CloudWatch metric anomaly detection
  3. CloudWatch custom metric
  4. CloudWatch dashboard

Explanation

Correct Answer

C:

Custom CloudWatch metrics record application-specific measurements not emitted by a service automatically; in this scenario, publish an explicit bad-record metric per run or dataset and alarm on a meaningful threshold.

Incorrect Answers

A:

Log-group retention settings automatically remove old log events after an assigned retention period; its benefit is controlling long-term storage of verbose ETL debug logs; that is different from monitoring per-batch invalid record counts.

B:

Anomaly detection models a metric's historical range and alarms on unusual deviations; the current need is for monitoring per-batch invalid record counts, not catching unexpected deviations from a periodic pipeline pattern.

D:

A dashboard displays chosen service and custom metrics in a shared operational view; this helps with correlating stream lag with transformation failures, not monitoring per-batch invalid record counts.

Question 80

A Python inventory program lists only the first thousand S3 objects and silently misses the rest. What SDK feature should be used?

  1. API response pagination contract
  2. ECS task role credential provider
  3. SDK client retry configuration
  4. Boto3 paginator

Explanation

Correct Answer

D:

Boto3 paginators iterate AWS API responses that return continuation tokens and limited pages; in this scenario, iterate all result pages with a paginator or continuation token; one API response is not necessarily the complete collection.

Incorrect Answers

A:

API consumers must follow service pagination and page-size contracts to avoid missing records; although useful for implementing a complete read of a partner's paginated REST feed, it leaves the need for retrieving every object across paginated AWS API responses unmet.

B:

The AWS SDK credential provider chain can obtain short-lived task-role credentials from an Amazon ECS task identity without static keys; it would target authenticating a scheduled ECS data-container AWS SDK client without static keys, while the question is about retrieving every object across paginated AWS API responses.

C:

AWS SDK retry settings manage retry attempts and backoff for supported retryable API failures; its benefit is handling transient AWS service throttling from ETL code; that is different from retrieving every object across paginated AWS API responses.

AWS Certified Data Engineer - Associate DEA-C01 Practice Tests