Data Ingestion and Transformation

1. A pipeline must ingest clickstream events at roughly 50,000 records per second, preserve ordering within each user session, and allow several independent consumers to read the same records. Which solution meets these requirements?

1. A Publish the events to an Amazon SNS topic and subscribe each application to the topic.
2. B Write the events to an Amazon Kinesis Data Stream partitioned by session identifier and run a separate consumer for each application.
3. C Write the events to an Amazon SQS standard queue and have each application poll the queue.
4. D Write the events to Amazon S3 and have each application list the bucket for new objects.

2. A Kinesis Data Streams consumer is falling behind and the iterator age is rising while each shard is at its read throughput limit. Which solution meets these requirements?

1. A Reduce the size of each record so more records fit within the shard limit.
2. B Increase the shard count so records are distributed across more shards, and scale the consumers accordingly.
3. C Increase the stream's retention period so records remain available for longer.
4. D Reduce the consumer's polling interval so records are read more frequently.

3. Several consumers read the same Kinesis Data Stream and compete for the shared per-shard read throughput. Which solution meets these requirements?

1. A Register the consumers to use enhanced fan-out so each receives dedicated throughput per shard.
2. B Increase the number of shards so the shared throughput is divided among more shards.
3. C Have each consumer poll less frequently so the shared throughput is not exhausted.
4. D Convert the stream to an Amazon Data Firehose delivery stream with multiple destinations.

4. An on-premises Oracle database must be replicated continuously into Amazon S3 with minimal downtime at cutover. Which solution meets these requirements?

1. A Create an AWS DMS task that performs a full load and then applies ongoing change data capture to Amazon S3.
2. B Create an AWS DMS task that performs a full load only, and run it during a maintenance window.
3. C Create an Amazon AppFlow flow that transfers the database tables to Amazon S3 on a schedule.
4. D Ship the database backups on AWS Snowball Edge devices and import them into Amazon S3.

5. Data must be ingested from Salesforce into Amazon S3 on a schedule without writing custom API integration code. Which solution meets these requirements?

1. A Create an AWS DMS task with Salesforce as the source endpoint and Amazon S3 as the target.
2. B Create an Amazon AppFlow flow with Salesforce as the source and Amazon S3 as the destination, on a scheduled trigger.
3. C Create an AWS Lambda function that polls the Salesforce API and writes the results to Amazon S3.
4. D Create an AWS Glue job that calls the Salesforce REST API and writes the results to Amazon S3.

6. An Amazon Data Firehose delivery stream must write records to Amazon S3, and the team must control how often a batch is delivered. Which combination of settings controls this? (Select TWO.)

1. A Set the number of shards on the source Kinesis Data Stream.
2. B Set the buffer interval in seconds so a batch is written when the interval elapses.
3. C Set the storage class of the destination Amazon S3 bucket.
4. D Set the version of the AWS Glue Data Catalog table used for conversion.
5. E Set the buffer size in megabytes so a batch is written when the buffer fills.

7. Records occasionally arrive twice in a streaming pipeline because the producer retries on timeout. Which solution meets these requirements?

1. A Modify the consumer to upsert on a deterministic record key so applying the same record twice produces the same result.
2. B Disable retries in the producer so a record is never sent more than once.
3. C Reduce the stream's retention period so duplicate records expire before they are processed.
4. D Add a DISTINCT clause to every query that reads the target table.

8. An AWS DataSync task copying an on-premises NFS share to Amazon S3 must verify that every transferred file matches the source. Which solution meets these requirements?

1. A Run the task a second time and confirm that no files are transferred again.
2. B Compare the object counts in the source and destination after the transfer.
3. C Configure the task's verification option to check data integrity after the transfer completes.
4. D Enable S3 Versioning on the destination bucket so each transfer is retained.

9. An ingestion job reading from an Amazon DynamoDB table is receiving throttling errors during peak periods. Which combination of steps meets these requirements? (Select TWO.)

1. A Convert the table to on-demand capacity mode, or enable auto scaling on its provisioned capacity.
2. B Increase the size of each item so fewer read requests are required.
3. C Implement exponential backoff with jitter in the reading client.
4. D Disable the SDK's built-in retry behaviour so failures surface immediately.
5. E Retry the throttled reads immediately in a loop until they succeed.

10. An existing Apache Kafka workload must move to AWS with minimal change to its producers and consumers. Which solution meets these requirements?

1. A Migrate the workload to Amazon EventBridge and route the messages to the existing consumers.
2. B Migrate the workload to Amazon Kinesis Data Streams and rewrite the producers and consumers.
3. C Migrate the workload to Amazon SQS and adapt the clients to the queue model.
4. D Migrate the workload to Amazon MSK so the existing Kafka clients connect with configuration changes only.

11. A streaming pipeline must invoke an AWS Lambda function for each batch of records read from an Amazon Kinesis Data Stream. Which solution meets these requirements?

1. A Write the stream records to Amazon S3 and invoke the function from an S3 event notification.
2. B Publish the stream records to an Amazon SNS topic and subscribe the function to the topic.
3. C Configure the Lambda function to poll the stream with the GetRecords API on a schedule.
4. D Create an event source mapping between the stream and the function, and configure the batch size and batching window.

12. An AWS Glue job must process only the files added to an Amazon S3 prefix since its previous run. Which solution meets these requirements?

1. A Increase the number of Glue workers so the full dataset is processed more quickly.
2. B Schedule an AWS Glue crawler to run immediately before each job execution.
3. C Add an AWS Glue Data Quality ruleset that filters previously processed records.
4. D Enable job bookmarks on the Glue job so processed state is persisted between runs.

13. A Spark job joins a 2 TB fact table with a 40 MB dimension table, and runtime is dominated by shuffling data between executors. Which solution meets these requirements?

1. A Increase the Amazon S3 read timeout so the shuffle has more time to complete.
2. B Write the output as CSV instead of Apache Parquet to reduce write time.
3. C Reduce the number of executors so less data is shuffled between them.
4. D Broadcast the dimension table to every executor so each join is performed locally.

14. An ETL script written in pandas processes a 3 GB dataset and does not require distributed processing. Which solution meets these requirements at the lowest cost?

1. A Run the script as an AWS Glue streaming job with a small window.
2. B Run the script on an Amazon EMR cluster with a single core node.
3. C Run the script as an AWS Glue Spark job with ten workers.
4. D Run the script as an AWS Glue Python shell job on a fraction of a DPU.

15. Source records arrive as deeply nested JSON, and analysts query only four top-level fields with filters on event date. Which combination of steps meets these requirements? (Select TWO.)

1. A Flatten the records to the required columns and write them in Apache Parquet.
2. B Partition the output by event date so queries prune unneeded prefixes.
3. C Load the records into a single unpartitioned CSV file for querying.
4. D Create an additional Amazon Athena workgroup for the analysts.
5. E Compress the JSON with gzip and leave the nested structure intact.

16. A data lake table receives frequent row-level updates and deletes, and analysts need consistent snapshot reads while writes are in progress. Which solution meets these requirements?

1. A Store the table as JSON files with a manifest listing the current file set.
2. B Store the table in an open table format such as Apache Iceberg so readers see a consistent snapshot.
3. C Store the table as CSV files partitioned by date and rewrite the affected partitions.
4. D Store the table as plain Apache Parquet files and refresh a Glue crawler after each write.

17. A pipeline must merge incoming change records into an existing dataset in Amazon S3, updating rows that exist and inserting those that do not. Which solution meets these requirements?

1. A Store the dataset in an open table format that supports a MERGE operation and apply the changes with it.
2. B Overwrite the entire dataset with the combined result on each run.
3. C Append all change records and resolve the latest version in every query.
4. D Run an AWS Glue crawler after each load so the catalog reflects the changes.

18. An AWS Glue job fails intermittently with an out-of-memory error on one executor while the others finish quickly. Which cause should the engineer investigate first?

1. A Investigate the Glue Data Catalog, because the table definition may be missing a partition.
2. B Investigate data skew, because one partition key is likely holding a disproportionate share of the records.
3. C Investigate the job's IAM role, because it may lack permission to read part of the data.
4. D Investigate the total worker count, because the job has insufficient parallelism overall.

19. Parquet files in a data lake must be compressed for frequent Amazon Athena queries where query speed matters more than the smallest file size. Which solution meets these requirements?

1. A Store the files uncompressed so no decompression is required at query time.
2. B Compress the files with Snappy, which decompresses quickly and remains splittable.
3. C Compress the files with gzip at the maximum compression level.
4. D Compress the files with bzip2 to achieve the smallest possible size.

20. A staging layer receives records appended row by row, and the source system adds fields every few weeks. Which file format is most appropriate?

1. A Write the staging data in Apache ORC for columnar compression.
2. B Write the staging data as fixed-width text for predictable parsing.
3. C Write the staging data in Apache Avro, a row-oriented format with schema evolution support.
4. D Write the staging data in Apache Parquet for columnar read efficiency.

21. An AWS Glue job must read from an on-premises PostgreSQL database over a private connection. Which combination of steps meets these requirements? (Select TWO.)

1. A Make the database publicly accessible and restrict it by source IP address.
2. B Embed the database credentials in the job script as parameters.
3. C Create a gateway VPC endpoint for Amazon S3 in the job's subnet.
4. D Store the database credentials in AWS Secrets Manager and grant the job role permission to retrieve them.
5. E Create an AWS Glue connection specifying the JDBC target, VPC subnet, and security group.

22. A transformation must mask email addresses and normalize inconsistent date formats, and the analysts performing the work prefer not to write code. Which solution meets these requirements?

1. A Write the transformation as an AWS Glue Spark script in Python.
2. B Build the transformation as an AWS Glue DataBrew recipe using its built-in masking and date transformations.
3. C Implement the transformation in an Amazon EMR notebook using Spark.
4. D Express the transformation as an Amazon Athena view over the source table.

23. An ETL job fails partway through and leaves partial output in the target Amazon S3 prefix, which downstream consumers then read. Which solution meets these requirements?

1. A Retry the job immediately whenever it fails so the output is completed sooner.
2. B Reduce the output file size so a failure affects fewer records.
3. C Increase the job timeout so the job is less likely to fail before completing.
4. D Write the output to a staging prefix and promote it to the target location only after the job completes successfully.

24. An Amazon EMR cluster must process a nightly batch that tolerates interruption, at the lowest cost that keeps the job reliable. Which solution meets these requirements?

1. A Run all nodes including the primary node on Spot capacity.
2. B Run the core nodes on On-Demand capacity for HDFS durability and the task nodes on Spot capacity.
3. C Run all nodes on On-Demand capacity to avoid any interruption.
4. D Run the job on a single large On-Demand instance outside of EMR.

25. A workflow must run an AWS Glue job, wait for it to succeed, then run two independent jobs in parallel, and alert on any failure. Which solution meets these requirements?

1. A Define the workflow as an AWS Step Functions state machine with a Parallel state and Catch handlers that publish to Amazon SNS.
2. B Send messages to an Amazon SQS queue between each stage so the jobs run in order.
3. C Create a Glue job that invokes the other jobs synchronously and reports failures in its logs.
4. D Create Amazon EventBridge rules that trigger each job when the previous job emits a success event.

26. A team must orchestrate data pipelines with existing Apache Airflow DAGs without operating the scheduler, workers, or web server. Which solution meets these requirements?

1. A Deploy the DAGs to an Amazon MWAA environment.
2. B Convert the DAGs into AWS Glue workflows.
3. C Add a Flink bootstrap action to an Amazon EMR cluster and run the DAGs on it.
4. D Install Apache Airflow on Amazon EC2 instances behind a load balancer.

27. An orchestration must fan out processing across 500 partitions in parallel and continue when a small number of partitions fail. Which solution meets these requirements?

1. A Use a Step Functions distributed Map state with a tolerated failure percentage and a configured concurrency limit.
2. B Use a Step Functions Parallel state with 500 branches defined in the state machine.
3. C Use a Step Functions Wait state between each partition so the workload is spread out.
4. D Use a Step Functions Choice state that iterates through the partitions in sequence.

28. A pipeline must start automatically when new objects land in an Amazon S3 prefix, and must not poll for them. Which solution meets these requirements?

1. A Create an Amazon MWAA sensor task that waits for objects to appear in the prefix.
2. B Create an AWS Lambda function that lists the prefix every minute and starts the pipeline when it finds new objects.
3. C Create an Amazon EventBridge rule on the S3 object created event that starts the pipeline execution.
4. D Create a scheduled AWS Glue trigger that runs the pipeline every hour.

29. A pipeline step must notify a downstream team when a stage completes, and the notification must reach both an email list and a queue for automated processing. Which solution meets these requirements?

1. A Send one completion event to each destination from the pipeline step in sequence.
2. B Publish the completion event to an Amazon SNS topic and subscribe both the email addresses and an Amazon SQS queue to it.
3. C Send the completion event to an Amazon SQS queue and have the email recipients poll the queue.
4. D Write the completion event to Amazon S3 and have both destinations read the object.

30. An AWS Lambda function in a data pipeline must be prevented from consuming the account's entire concurrency pool during a spike. Which solution meets these requirements?

1. A Increase the function's timeout so invocations are less likely to fail under load.
2. B Configure reserved concurrency on the function so its concurrent executions are capped.
3. C Configure provisioned concurrency on the function so environments are pre-initialized.
4. D Increase the function's memory allocation so each invocation completes more quickly.

31. A serverless data pipeline of Lambda functions, a Step Functions state machine, and a DynamoDB table must be packaged for repeatable deployment. Which solution meets these requirements?

1. A Create the resources manually and generate a CloudFormation template from them afterwards.
2. B Create each resource in the AWS Management Console and record the settings in a document.
3. C Write a shell script that calls the AWS CLI to create each resource in sequence.
4. D Define the resources in an AWS SAM template and deploy them with the AWS SAM CLI.

32. An Amazon Redshift query that aggregates a large fact table runs slowly, and the execution plan shows a full scan of every partition. Which solution meets these requirements?

1. A Increase the cluster's node count so the scan completes more quickly.
2. B Define a sort key on the column used in the query's range filter so zone maps can skip blocks.
3. C Convert the query to use SELECT * so the planner has full column statistics.
4. D Increase the query timeout so the scan has time to complete.

33. Transformation logic that runs inside Amazon Redshift must be reusable, parameterized, and callable from several pipelines. Which solution meets these requirements?

1. A Create a Redshift stored procedure that encapsulates the logic and call it from each pipeline.
2. B Implement the logic in an AWS Lambda function that issues the SQL statements.
3. C Create an Amazon Athena view over the Redshift tables and query it from each pipeline.
4. D Copy the SQL statements into each pipeline's job definition.

34. A data engineering team must manage pipeline code in source control with feature branches merged after review. Which set of Git actions supports this workflow?

1. A Copy the repository directory, edit the copy, and replace the original when finished.
2. B Clone the repository and commit all changes directly to the main branch.
3. C Create a new repository for each change and archive the previous one.
4. D Clone the repository, create a feature branch, commit changes to it, and open a request to merge the branch after review.

35. An AWS Lambda function in a pipeline must write temporary files larger than its ephemeral storage allows. Which solution meets these requirements?

1. A Increase the function's memory allocation so more temporary space is available.
2. B Increase the function's timeout so the files can be written in smaller pieces.
3. C Mount an Amazon EFS access point in the function and write the temporary files to the file system.
4. D Write the temporary files to the function's /tmp directory and delete them between invocations.

36. A Kinesis Data Streams producer is receiving throughput exceeded errors on one shard while other shards are underused. Which cause should be investigated first?

1. A Investigate whether the partition key concentrates records onto that shard, since throughput limits apply per shard.
2. B Investigate whether the consumer is reading too slowly.
3. C Investigate whether the stream is in on-demand capacity mode.
4. D Investigate whether the stream's retention period is too short.

37. An AWS DMS task performing ongoing replication is falling behind the source database. Which combination of steps meets these requirements? (Select TWO.)

1. A Restart the task from the beginning to reset the lag.
2. B Convert the task to full load only so the backlog is cleared.
3. C Check the replication instance's resource utilization and size it up if CPU or memory is saturated.
4. D Review the source database's transaction log retention so changes are not purged before DMS reads them.
5. E Reduce the target's write capacity so the task consumes less.

38. An Amazon Data Firehose delivery stream must transform records before writing them to Amazon S3. Which solution meets these requirements?

1. A Configure a Lambda transformation function on the delivery stream so each batch is processed in flight.
2. B Write the records to S3 and run an AWS Glue job on the bucket afterwards.
3. C Configure the producer to transform records before sending them.
4. D Enable dynamic partitioning on the delivery stream.

39. An ingestion pipeline must detect duplicate records arriving from a source that retries on timeout. Which approach is appropriate?

1. A Compare each incoming record against every existing record.
2. B Ignore duplicates because they are rare.
3. C Assign each record a deterministic identifier at the source and deduplicate on it at ingestion.
4. D Reduce the source's timeout so retries are less frequent.

40. A batch ingestion job reads from an API that limits requests per second. Which approach is appropriate?

1. A Throttle the job's request rate to the limit and retry throttled calls with exponential backoff.
2. B Retry throttled calls immediately until they succeed.
3. C Request all the data in one large call.
4. D Run several copies of the job in parallel to finish faster.

41. An Amazon MSK cluster's consumers must resume from where they stopped after a restart. Which mechanism provides this?

1. A The topic's replication factor.
2. B The producer's acknowledgement setting.
3. C The cluster's retention period, which keeps records available.
4. D Consumer group offsets committed to the cluster, which record each partition's last processed position.

42. A Glue job's output has many partitions containing a handful of tiny files each. Which cause should be investigated first?

1. A Investigate whether the job bookmark is enabled.
2. B Investigate whether the job writes with high parallelism into a partition key with high cardinality, producing one small file per task per partition.
3. C Investigate whether the job's worker type is too small.
4. D Investigate whether the source data is compressed.

43. A transformation must produce a slowly changing dimension with history preserved. Which write pattern is appropriate?

1. A Append a new row without closing the previous one.
2. B Delete the row and insert the new value.
3. C Close the current row by setting its end date and insert a new row for the changed value.
4. D Overwrite the row in place with the new value.

44. A Spark job must join a 5 TB table with a 3 TB table on a key that is evenly distributed. Which approach is appropriate?

1. A Allow a shuffle join, since neither side fits in memory for a broadcast.
2. B Collect the 5 TB table to the driver and join there.
3. C Broadcast the 3 TB table to every executor.
4. D Convert both tables to a single partition before joining.

45. A Glue job must apply a transformation that uses a Python library not included in the Glue runtime. Which approach is appropriate?

1. A Install the library on the Glue workers at the start of each run.
2. B Package the library and reference it through the job's additional Python modules parameter.
3. C Rewrite the transformation without the library.
4. D Run the transformation on Amazon EMR instead.

46. Records in a data lake table must be updated in place without rewriting whole partitions. Which table format provides this?

1. A Apache Parquet files partitioned by date.
2. B JSON files rewritten on each update.
3. C An open table format such as Apache Iceberg or Apache Hudi that supports row-level updates.
4. D CSV files with a manifest listing the current set.

47. A transformation must handle a source column whose type changed from integer to string mid-way through the historical data. Which approach is appropriate?

1. A Cast the column to a common type during the transformation so both eras conform to one schema.
2. B Store the column as two separate columns.
3. C Reject the dataset until the source corrects the history.
4. D Drop the rows from before the change.

48. A Step Functions workflow must retry a Glue job on transient failure but stop after three attempts. Which configuration is appropriate?

1. A A Retry field on the task state specifying the error types, interval, and maximum attempts.
2. B A Wait state before the task so transient conditions clear.
3. C A Parallel state running three copies of the job.
4. D A Catch field routing all errors to a failure state.

49. An Airflow DAG on Amazon MWAA must not start a new run while the previous run is still executing. Which configuration is appropriate?

1. A Add a sensor task at the start of the DAG.
2. B Set the DAG's maximum active runs to one so overlapping runs are prevented.
3. C Reduce the number of workers in the environment.
4. D Increase the DAG's schedule interval.

50. A pipeline must run daily but skip execution when the input for that day has not arrived. Which approach is appropriate?

1. A Wait indefinitely for the input to arrive.
2. B Fail the workflow when the input is absent so the team is alerted.
3. C Process whatever data exists regardless of completeness.
4. D Check for the input at the start of the workflow and end successfully without processing if it is absent.

51. A workflow step depends on the output of two independent upstream steps. Which Step Functions structure is appropriate?

1. A Two separate state machines with the dependent step in a third.
2. B A Parallel state containing the two upstream steps, followed by the dependent step.
3. C Three sequential task states.
4. D A Map state iterating over the two upstream steps.

52. A Lambda function in a pipeline must process a batch from a queue and must not lose records if it fails partway. Which approach is appropriate?

1. A Process one record per invocation.
2. B Process the whole batch and delete it on completion.
3. C Report partial batch failures so only the unprocessed records are returned to the queue.
4. D Increase the function's timeout so the batch completes.

53. A SQL transformation must compute a running total ordered by date within each account. Which construct is appropriate?

1. A A subquery computing the total per account.
2. B A window function with a SUM over a frame partitioned by account and ordered by date.
3. C A self join on account and date.
4. D A GROUP BY on account with SUM.

54. A pipeline's infrastructure must be deployed identically to development and production accounts. Which approach is appropriate?

1. A Create the resources in the console in each account following a checklist.
2. B Define the infrastructure as code and deploy the same template to each account with environment-specific parameters.
3. C Export the development resources and import them into production.
4. D Share a single set of resources between both environments.

55. A data engineer must query an Amazon Redshift cluster from a Lambda function without managing database connections. Which approach is appropriate?

1. A Route the query through Amazon Athena.
2. B Maintain a connection pool inside the function.
3. C Use the Redshift Data API, which issues SQL asynchronously through an HTTPS endpoint.
4. D Open a JDBC connection from the function on each invocation.

56. A pipeline's code must be reviewed before it reaches the main branch. Which Git workflow supports this?

1. A Develop on a feature branch and merge through a reviewed pull request.
2. B Tag each commit with the reviewer's name.
3. C Copy the repository and edit the copy.
4. D Commit directly to the main branch and review afterwards.

57. A Kinesis Data Streams application must read the same records as another consumer without sharing read throughput. Which solution meets these requirements?

1. A Increase the stream's retention period.
2. B Register the application as an enhanced fan-out consumer so it receives dedicated throughput per shard.
3. C Have the second consumer read from a copy of the stream.
4. D Add shards to the stream so the shared throughput is divided further.

58. An AWS DMS migration must validate that the target data matches the source after the load. Which configuration is appropriate?

1. A Run the load twice and compare the results.
2. B Enable data validation on the task so DMS compares source and target rows and reports mismatches.
3. C Compare row counts manually after the load.
4. D Enable CloudWatch logging on the replication instance.

59. An ingestion pipeline must read from an on-premises Kafka cluster during a phased migration. Which approach is appropriate?

1. A Rebuild the producers to write to both clusters simultaneously.
2. B Have consumers in AWS connect directly to the on-premises cluster over the internet.
3. C Use MSK Connect or a mirroring tool to replicate topics into Amazon MSK while producers migrate.
4. D Export the topics to files and upload them daily.

60. A Firehose delivery stream must write to Amazon S3 in Parquet using a schema from the Data Catalog. Which configuration is required?

1. A Configure the producer to send Parquet directly.
2. B Enable record format conversion on the delivery stream and reference the Glue Data Catalog table.
3. C Enable compression on the delivery stream.
4. D Write JSON to S3 and convert it with a scheduled Glue job.

61. A streaming ingestion pipeline must not lose records if the destination is briefly unavailable. Which configuration is appropriate?

1. A Configure a backup destination for records that fail delivery so they can be reprocessed.
2. B Increase the destination's capacity.
3. C Reduce the producer's send rate.
4. D Increase the delivery stream's buffer interval.

62. A Glue job must process a dataset that grows each day without reprocessing prior days. Which configuration is appropriate?

1. A Reduce the job's worker count to limit throughput.
2. B Filter the input by date inside the job's script.
3. C Enable job bookmarks so the job tracks what it has already processed between runs.
4. D Delete processed files after each run.

63. A Spark transformation produces one output file per partition, and downstream readers require files of a consistent size. Which approach is appropriate?

1. A Write the output uncompressed.
2. B Reduce the number of partitions in the source data.
3. C Repartition or coalesce the data before writing so file sizes are controlled.
4. D Increase the number of executors so more files are written.

64. A pipeline must convert a nested JSON structure into a flat table for analytical queries. Which transformation is appropriate?

1. A Compress the JSON before loading it.
2. B Store the JSON as a single string column.
3. C Flatten or relationalize the nested structure so each nested field becomes a column or a related table.
4. D Increase the query engine's memory allocation.

65. A transformation must deduplicate records while keeping the most recent version of each key. Which approach is appropriate?

1. A Delete all but the first record encountered for each key.
2. B Apply DISTINCT to the whole record.
3. C Rank the records by timestamp within each key using a window function and keep the highest ranked.
4. D Group by the key and take the maximum of every column.

66. A Glue job's transformation logic must be unit tested before deployment. Which approach is appropriate?

1. A Review the script in a pull request without executing it.
2. B Run the job against production data with a small limit.
3. C Deploy the job and inspect its output.
4. D Structure the transformation as testable functions and run them against sample data locally or in a test harness.

67. A pipeline must process several partitions concurrently with a bounded number in flight. Which Step Functions construct is appropriate?

1. A A Parallel state with a branch per partition.
2. B A sequence of task states, one per partition.
3. C A Choice state iterating through the partitions.
4. D A Map state with a concurrency limit so partitions are processed in parallel up to the bound.

68. A workflow must wait for a Glue job to complete before starting the next step. Which integration pattern is appropriate?

1. A Invoke the job and immediately proceed to the next state.
2. B Poll the job's status from a Lambda function in a loop.
3. C Invoke the job and add a fixed wait state sized to the expected duration.
4. D Invoke the job with the run-a-job integration pattern so the state waits for completion.

69. An orchestration must record which input files each pipeline run processed. Which approach is appropriate?

1. A Record the number of files processed.
2. B Record the execution start time only.
3. C Pass the input manifest through the workflow and record it with the execution, so the run is traceable to its inputs.
4. D Rely on the S3 bucket's object listing at review time.

70. A Lambda function processing a Kinesis stream must not block the shard when one record repeatedly fails. Which configuration is appropriate?

1. A Reduce the event source mapping's batch size to one.
2. B Configure a maximum retry attempt count and an on-failure destination so the problematic record is set aside.
3. C Increase the function's timeout so processing completes.
4. D Increase the function's memory allocation.

71. A SQL query must return the difference between each row's value and the previous row's value ordered by date. Which construct is appropriate?

1. A A self join on a date offset.
2. B A subquery computing the overall average.
3. C A GROUP BY on the date column.
4. D A window function using LAG over the ordered rows.

72. A data pipeline's Python code must be shared across several Glue jobs without duplication. Which approach is appropriate?

1. A Copy the code into each job's script.
2. B Store the code in a database each job reads at runtime.
3. C Package the shared code as a library in Amazon S3 or CodeArtifact and reference it from each job.
4. D Rewrite the shared logic in SQL for each job.

73. A pipeline must ingest data from a SaaS application on a schedule without writing API integration code. Which solution meets these requirements?

1. A Export the data manually and upload it.
2. B Write a Lambda function that calls the application's API.
3. C Create a Glue job that reads the application's REST endpoints.
4. D Create an Amazon AppFlow flow with the SaaS application as the source and a scheduled trigger.

74. An ingestion job must read a compressed file format that cannot be split across workers. Which consequence applies?

1. A The file is read faster than a splittable format.
2. B The file must be converted to JSON before reading.
3. C The file cannot be read at all.
4. D The file is processed by a single worker, so large files of that format limit parallelism.

75. A streaming source produces records faster than the configured shard count can accept. Which symptom appears first?

1. A The stream's retention period shortens.
2. B Producers receive throughput exceeded errors on the affected shards.
3. C Consumers report a rising iterator age.
4. D Records are silently discarded.

76. An ingestion pipeline must preserve the original raw files alongside the processed output. Which approach is appropriate?

1. A Overwrite the raw files with the processed output.
2. B Store raw and processed data in the same prefix.
3. C Delete the raw files once processing succeeds.
4. D Write the raw files to a separate landing prefix with its own lifecycle policy, and write processed output elsewhere.

77. A DMS task must migrate a source table that has no primary key. Which consideration applies?

1. A The target table must use the same engine as the source.
2. B Ongoing replication of updates and deletes requires a way to identify rows uniquely, so a key or unique index is needed.
3. C The task must run in full load mode only, which is always acceptable.
4. D The table cannot be migrated at all.

78. An ingestion pipeline must handle a source that occasionally sends records out of order. Which approach is appropriate?

1. A Include an event timestamp in each record and order by it during processing rather than relying on arrival order.
2. B Buffer all records until the batch is complete.
3. C Increase the stream's retention period.
4. D Reject records that arrive out of order.

79. A pipeline must ingest data from a database without affecting the source system's performance. Which approach is appropriate?

1. A Read from a read replica or use log-based change data capture rather than querying the primary directly.
2. B Increase the primary's instance size.
3. C Query the primary with a longer timeout.
4. D Query the primary during business hours with a row limit.

80. An ingestion pipeline must confirm that a delivered file is complete before processing begins. Which approach is appropriate?

1. A Wait a fixed period after the file appears.
2. B Check the file's size against an expected value.
3. C Have the producer write a marker file after the data file, and trigger processing on the marker.
4. D Trigger processing on the data file's creation event.

81. A Lambda function processing pipeline events must handle a downstream service being temporarily unavailable. Which approach is appropriate?

1. A Retry immediately until the service responds.
2. B Retry with exponential backoff and route persistent failures to a dead-letter destination for reprocessing.
3. C Log the failure and discard the event.
4. D Increase the function's timeout.

82. A SQL transformation must assign a sequential number to rows within each group ordered by date. Which construct is appropriate?

1. A A self join counting earlier rows.
2. B A subquery returning the group's total.
3. C COUNT with a GROUP BY on the group.
4. D ROW_NUMBER as a window function partitioned by the group and ordered by date.

83. A pipeline's Python code must handle a schema field that is sometimes absent. Which approach is appropriate?

1. A Reject records missing the field.
2. B Assume the field is always present.
3. C Convert every field to a string.
4. D Access the field with a default so an absent field yields a known value rather than raising an error.

84. A data pipeline's SQL must be version controlled and reviewed like application code. Which approach is appropriate?

1. A Store the SQL in the repository and apply it through the deployment pipeline rather than editing it in a console.
2. B Edit the SQL in the query console and record the change in a ticket.
3. C Store the SQL in a shared document.
4. D Keep the SQL in each engineer's local files.

85. A transformation must convert a string column containing dates in several formats into a date type. Which approach is appropriate?

1. A Convert every value to the current date when parsing fails.
2. B Parse with a single format and drop failures.
3. C Store the column as a string and parse at query time.
4. D Parse each value against the known formats in order, routing values matching none to quarantine.

86. A pipeline's transformation logic must be shared between a batch job and a streaming job. Which approach is appropriate?

1. A Extract the logic into a shared library both jobs import so behaviour stays identical.
2. B Convert the streaming job to batch.
3. C Implement the logic separately in each job.
4. D Run the batch job on the streaming output.

87. A Redshift stored procedure must handle an error without aborting the entire transaction. Which construct is appropriate?

1. A An exception handling block that catches the error and takes a defined action.
2. B A rollback at the start of the procedure.
3. C A longer statement timeout.
4. D A commit after every statement.

88. A transformation must produce the same result when it is re-run over the same input. Which property is required?

1. A Compression of the output files
2. B Determinism, so the transformation does not depend on the time it runs or on unordered input
3. C Partitioning of the input data
4. D Idempotency of the destination table