AWS Certified Data Engineer - Associate (DEA-C01) Practice Exam

What the AWS Certified Data Engineer - Associate (DEA-C01) exam covers

• Data Ingestion and Transformation273 questions
• Data Store Management209 questions
• Data Operations and Support175 questions
• Data Security and Governance145 questions

Free AWS Certified Data Engineer - Associate (DEA-C01) sample questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 802.

1. Question 1Data Ingestion and Transformation

   An analyst wants near-real-time indexing of application logs into a search cluster for operational dashboards, tolerating a delivery latency of about one minute. Which Firehose destination configuration is designed for this?

   • AOpenSearch Service as the destinationCorrect
   • BAmazon S3 with Parquet conversion enabled
   • CAmazon Redshift as the destination
   • DAn HTTP endpoint to a batch reporting tool
   ✓ Correct answer: A

   Firehose has a native OpenSearch Service destination that indexes incoming records into a specified domain and index, which is well suited to operational log analytics dashboards that need data available within roughly a minute rather than requiring true sub-second streaming.

   Why the other options are wrong

   • BS3 with Parquet conversion is intended for cost-efficient analytical storage and batch querying, not live search dashboards.
   • CRedshift is optimized for structured analytical queries via SQL, not for the search-style indexing needed by log dashboards.
   • DAn HTTP endpoint to a batch reporting tool implies periodic batch consumption rather than near-real-time search indexing.
2. Question 2Data Ingestion and Transformation

   A company runs a low volume Glue streaming ETL job that only needs a small amount of compute per worker, and wants to minimize cost without switching to a non-Spark job type. Which worker type is designed for this scenario?

   • AG.025X workers sized for small, low cost workloadsCorrect
   • BG.1X workers sized for general purpose workloads
   • CG.4X workers sized for the largest workloads
   • DG.2X workers sized for more memory intensive workloads
   ✓ Correct answer: A

   G.025X workers provide a smaller compute and memory footprint than a standard G.1X worker and are intended for low volume streaming ETL jobs, letting them run more cheaply while still using Spark structured streaming and DynamicFrame processing, unlike switching to a Python shell job which cannot run streaming Spark workloads at all. G.1X, G.2X, and G.4X workers provide progressively more compute and memory for general purpose, memory intensive, and the largest workloads respectively.

   Why the other options are wrong

   • BG.1X workers are the general purpose default size and cost more than the smaller G.025X option.
   • CG.4X workers are sized for the largest and most demanding Spark workloads, not small low volume jobs.
   • DG.2X workers add extra memory for larger, more demanding jobs, which increases cost unnecessarily here.
3. Question 3Data Ingestion and Transformation

   After a Lambda task succeeds, a workflow must immediately run two independent downstream tasks, a Redshift load and an S3 archive copy, that do not depend on each other, and total execution time should be minimized. Which construct should follow the Lambda task?

   • AA Choice state that randomly selects one task to run
   • BA Parallel state with one branch for each taskCorrect
   • CA Map state iterating twice over the same input
   • DTwo sequential Task states, load then archive copy
   ✓ Correct answer: B

   A Parallel state runs multiple branches of independent logic concurrently within one state, so placing the Redshift load in one branch and the archive copy in another lets both run at the same time rather than waiting on each other. Two sequential Task states would needlessly serialize work that has no real dependency.

   Why the other options are wrong

   • AA Choice state picks one branch based on a condition and would skip the other task entirely, not run both.
   • CA Map state repeats the same logic over a collection of items, not two different unrelated tasks.
   • DSequential states force the archive copy to wait for the load to finish despite no actual dependency, increasing runtime.
4. Question 4Data Ingestion and TransformationSelect all that apply

   A stream is experiencing a hot shard because a small number of partition key values receive disproportionately high traffic. Which two actions would help resolve this issue? (Choose two.)

   • ASplit the affected shard so hot traffic spreads across more shardsCorrect
   • BReduce the stream's data retention period down to 24 hours
   • CRedesign the partition key scheme to use a higher cardinality keyCorrect
   • DDisable enhanced fan-out for every registered consumer on the stream
   • ESwitch the KCL consumer application to run with a single threaded worker
   ✓ Correct answer: A, C

   Choosing partition key values with higher cardinality and less skew spreads records across more shards, and splitting the specific overloaded shard divides its hash key range so the hot traffic is shared by multiple shards going forward. Neither retention changes, consumer fan-out settings, nor consumer threading affect how write traffic is distributed across shards.

   Why the other options are wrong

   • BRetention period controls how long data is stored and has no effect on write distribution across shards.
   • DEnhanced fan-out is a read side feature and does not influence how writes are distributed across shards.
   • EConsumer threading affects read side processing, not how producer traffic is hashed onto shards.
5. Question 5Data Store Management

   A web application stores session tokens in DynamoDB and needs expired sessions to be automatically removed from the table without writing a scheduled cleanup job. Which DynamoDB feature should be used?

   • AA global secondary index on expiration_date
   • BThe DynamoDB TTL attribute for expirationCorrect
   • CDynamoDB Streams for change data capture
   • DPoint-in-time recovery for backup and restore
   ✓ Correct answer: B

   By marking an attribute as the TTL attribute and storing an expiration timestamp on each item, DynamoDB's background process automatically identifies and deletes expired items without any custom scheduled job or additional application logic. This is well suited to use cases like session token expiry where stale items should be purged automatically.

   Why the other options are wrong

   • AA GSI helps query by expiration_date but still requires a process to delete items.
   • CStreams captures change events but does not itself delete expired items.
   • DPoint-in-time recovery is a backup and restore capability, not an expiration mechanism.
6. Question 6Data Store Management

   A data engineer wants to run interactive SQL queries against Parquet files in Amazon S3 without provisioning or managing any compute clusters. Which service fits this requirement?

   • AAmazon EMR, which needs EC2 nodes launched before jobs can run
   • BAmazon Athena, which queries S3 with no clusters to provisionCorrect
   • CAWS Glue ETL jobs, which run scripted batch transforms, not SQL
   • DAmazon Redshift, needing a cluster provisioned before queries run
   ✓ Correct answer: B

   Athena is built on the open source Trino (formerly Presto) distributed query engine and requires no infrastructure provisioning; users submit SQL and pay per query based on data scanned. The other options all require sizing, launching, or managing compute infrastructure before queries or jobs can run.

   Why the other options are wrong

   • AEMR requires launching and managing EC2-based cluster nodes for processing.
   • CGlue ETL jobs run scripted transformations and are not meant for interactive ad hoc SQL.
   • DRedshift needs a provisioned cluster sized in advance, unlike Athena's serverless model.
7. Question 7Data Store ManagementSelect all that apply

   Which two factors would lead a team to choose Apache Iceberg over plain Parquet files for a table on S3? (Choose two.)

   • AThe workload issues frequent row-level updates and deletes against existing recordsCorrect
   • BThe workload only ever appends new rows and never updates or deletes existing ones
   • CThe team wants the smallest possible per-file compression ratio for the raw data
   • DThe team never expects the table's schema to change after initial creation
   • EThe team needs to query the table as it existed at a specific point in time using snapshotsCorrect
   ✓ Correct answer: A, E

   Iceberg's metadata layer tracks row-level changes and historical snapshots, which directly supports frequent updates and deletes without full-partition rewrites, and lets queries target the table's state as of a prior snapshot. An append-only workload, a focus on compression ratio, or a schema that never changes are all scenarios plain Parquet already handles well, so they do not by themselves justify adopting Iceberg's additional table format overhead.

   Why the other options are wrong

   • BAn append-only workload with no updates or deletes does not need Iceberg's row-level change tracking; plain Parquet already handles it well.
   • CPer-file compression ratio depends on the underlying file format such as Parquet, not on whether Iceberg's table format is layered on top.
   • DA schema that never changes does not benefit from Iceberg's schema evolution capability, since there is nothing to evolve.
8. Question 8Data Operations and Support

   A Redshift administrator can either set a WLM query timeout in seconds or create a query monitoring rule with a metric-based predicate. What key advantage does the query monitoring rule have over a simple WLM timeout?

   • AIt can trigger on metrics such as rows scanned or CPU timeCorrect
   • BIt only applies to queries submitted through the JDBC driver
   • CIt permanently blocks the offending user from connecting again
   • DIt automatically upgrades the cluster's node type when triggered
   ✓ Correct answer: A

   A WLM timeout only measures elapsed time, whereas a query monitoring rule predicate can be based on metrics such as rows scanned, CPU time, or return row count, giving much finer control over which runaway behavior triggers an action. This lets administrators target specific resource-consumption patterns rather than only duration.

   Why the other options are wrong

   • BQuery monitoring rule predicates apply to queries regardless of which client driver submitted them.
   • CQuery monitoring rule actions never revoke a user's connection permission permanently, they act on the query itself.
   • DQuery monitoring rules cannot resize the cluster or change node types.
9. Question 9Data Operations and Support

   A data warehouse uses a star schema where fact records sometimes arrive before their corresponding dimension record has been loaded, a situation commonly called a late-arriving dimension. What is the standard approach to handle this in the fact table load?

   • AInsert a placeholder dimension row that gets updated once the real record arrivesCorrect
   • BReject the fact record entirely until the dimension record is confirmed to exist
   • CStore the fact record with a null foreign key until a nightly batch job resolves it
   • DDuplicate the fact record into a separate table reserved for orphaned records
   ✓ Correct answer: A

   Inserting an inferred member row, keyed by the natural key with default or unknown attribute values, lets the fact record load immediately with a valid surrogate key reference; the placeholder row is later updated in place once the true dimension attributes are loaded, preserving referential integrity throughout.

   Why the other options are wrong

   • BRejecting the fact record delays visibility of valid business events and is not the standard pattern for this scenario.
   • CUsing a null foreign key breaks the star schema's referential integrity and complicates joins until resolved.
   • DRouting to a separate orphan table adds unnecessary complexity when a placeholder dimension row handles the case directly within the existing schema.
10. Question 10Data Security and Governance

    A team wants to reuse one Lake Formation data filter across several different tables. Which statement is correct?

    • AA single data filter can be attached to grants for many different tables at the same time.
    • BData filters are account-wide objects that automatically apply to every table in the catalog.
    • CEach data filter is scoped to one table, so other tables each need their own data filter.Correct
    • DA data filter automatically applies to all tables that happen to share the same column names.
    ✓ Correct answer: C

    Each data filter is created in the context of a single table's schema, since its column list and row-filter expression are evaluated against that table's columns. To apply equivalent row or column restrictions to another table, a separate data filter must be created, even when the restriction logic is similar.

    Why the other options are wrong

    • AA data filter cannot be attached to multiple tables at once.
    • BData filters are per-table, not account-wide automatic objects.
    • DMatching column names does not make one filter span multiple tables.

Related AWS resources

• AWS Certified Data Engineer - Associate (DEA-C01) study guideKey concepts
• AWS practice examsAll AWS
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• AWS SCS-C03 practice examRelated
• AWS Certified AI Practitioner (AIF-C01) practice examRelated
• AWS CLF-C02 practice examRelated

AWS Certified Data Engineer - Associate (DEA-C01) practice exam FAQ