15 Free AWS Certified Data Engineer – Associate (DEA-C01) Practice Questions with Answers (2026)

The partition key decides which shard a record goes to. One key sends every record to one shard, which is capped at 1 MB or 1,000 records per second for writes. A key such as the device ID spreads records across all shards.

• B is wrong: with a single key, every record still goes to one shard, so more shards do not help.
• C is wrong: enhanced fan-out improves read throughput for consumers, not write throughput for producers.
• D is wrong: retention controls how long records are kept, not how fast they can be written.

Step Functions can start a Glue job and wait for it to complete, invoke Lambda, and define retries, backoff and catch blocks for errors. It is serverless and pay-per-use.

• A is wrong: Glue workflows orchestrate Glue jobs and crawlers but cannot run a Lambda function as a step.
• B is wrong: an EC2 cron job is a server you must manage, patch and monitor.
• C is wrong: fixed schedules do not wait for the previous step to finish and have no retry or error path.

The parallelization factor lets Lambda process up to 10 batches from each shard at the same time. Lambda still keeps records with the same partition key in order.

• A is wrong: longer retention keeps data from expiring but does not make processing faster.
• C is wrong: a standard queue does not guarantee order.
• D is wrong: smaller batches mean more invocations and more overhead, so the backlog grows.

AppFlow is a managed integration service with built-in connectors for software-as-a-service apps such as Salesforce. It can run flows on a schedule and write to S3.

• A is wrong: DMS migrates data from databases, not from Salesforce.
• B is wrong: Firehose delivers streaming data that a producer sends to it. It does not pull from Salesforce.
• D is wrong: Transfer Family provides SFTP, FTPS and FTP endpoints for file transfers, not a Salesforce connector.

COPY loads data in parallel across the cluster's slices. Splitting the input into several compressed files lets every slice work at once.

• A is wrong: single-row inserts are very slow for bulk loads.
• C is wrong: Spectrum queries data where it sits in S3. It does not load it into Redshift tables.
• D is wrong: UNLOAD exports data from Redshift to S3, the opposite direction.

Deep Archive is the lowest-cost storage class, and its standard retrieval completes within 12 hours. An expiration action deletes the logs at 7 years, as compliance requires.

• A is wrong: there is no deletion, which breaks the 7-year rule, and it costs more than Deep Archive for data almost never read.
• B is wrong: Standard-IA costs far more than Deep Archive for data that is almost never read.
• C is wrong: the logs are never deleted, and Flexible Retrieval costs more than Deep Archive.

Athena only reads partitions it knows about. New prefixes must be registered in the Data Catalog, or partition projection must be set up so Athena works them out from the query.

• B is wrong: the file format does not stop Athena from seeing a partition.
• C is wrong: workgroup limits control query concurrency and data scanned, not which partitions are visible.
• D is wrong: versioning keeps old object versions and has nothing to do with partitions.

TTL deletes expired items automatically in the background, and the deletes do not consume write capacity. Deletion usually happens within a few days of expiry, so filter out expired items in reads if timing matters.

• A is wrong: scans and deletes consume read and write capacity and need code to maintain.
• B is wrong: S3 Lifecycle rules act on S3 objects, not on DynamoDB items.
• D is wrong: Streams record changes to items. They do not delete anything.

Glue Data Quality lets you define rules, such as a completeness threshold on a column, and evaluate them inside the job. A failed rule can stop the job before bad data is written.

• A is wrong: CloudTrail records API activity, not data contents.
• C is wrong: checking afterwards means the bad data is already in the target.
• D is wrong: Macie finds sensitive data such as PII. It does not check data quality rules.

Glue sends job state change events to EventBridge. A rule matching the FAILED state can notify an SNS topic straight away.

• A is wrong: a daily query is not as soon as it fails.
• B is wrong: crawlers update the Data Catalog and do not send alerts.
• C is wrong: a manual morning check is not an alert.

CloudTrail records management API calls, including who made them. Event history keeps the last 90 days of management events, so a DeleteTable call from yesterday is there.

• B is wrong: metrics show counts and performance, not who made a call.
• C is wrong: S3 access logs record requests to the bucket, not Data Catalog API calls.
• D is wrong: flow logs record network traffic, not API calls.

Lake Formation supports column-level permissions on Data Catalog tables, and Athena enforces them. One copy of the data serves every group.

• A is wrong: a second copy has to be kept in sync and doubles storage.
• C is wrong: bucket policies work on objects, not on columns inside files.
• D is wrong: IAM has no condition keys for individual table columns.

Macie uses managed data identifiers and machine learning to discover sensitive data, including PII, in S3.

• A is wrong: GuardDuty detects threats from activity, not sensitive data inside objects.
• B is wrong: Inspector scans workloads for software vulnerabilities.
• D is wrong: Config tracks resource configuration, not object contents.

Cross-account use of a KMS key needs permission on both sides: the key policy in the owning account must allow the external role, and the role's own policy must allow the action on that key.

• B is wrong: the requirement is to keep using the customer managed key.
• C is wrong: versioning has nothing to do with key permissions.
• D is wrong: you cannot copy a KMS key or export its key material into another account.