Amazon AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 27

Question 260

A public REST API fronts a third-party service limited to 100 requests per second. Clients may burst well above that. Which gateway behavior prevents overload?

Set all requests to unlimited concurrency and rely only on the third party returning 429.
Set a CloudFront response cache TTL for API GET requests and assume this imposes a strict global 100-call-per-second ceiling on all uncached API methods.
Configure API Gateway token-bucket throttling/usage plans as protective targets, and add a downstream concurrency limiter or queue if the third-party 100-RPS ceiling is strict.
Increase the S3 object retention period for the API host.

Explanation

Correct Answer

API Gateway throttles and quotas lower the flow of bursts to an upstream dependency, but AWS documents those limits as best-effort targets, not guaranteed hard ceilings. Protect a truly strict contract with an application-side token bucket, bounded integration concurrency or durable queue and careful backoff.

Incorrect Answers

This overloads the dependency and increases failures before protective controls act.

Caching may reduce origin requests for eligible responses, but it does not impose a reliable requests-per-second quota on all uncached API operations. Use throttling plus downstream protection.

Storage retention cannot limit backend request throughput.

Question 261

A DynamoDB Orders table update must asynchronously update a search index without delaying the transaction that writes the table. Which DynamoDB integration supplies a change stream?

Enable DynamoDB Streams and configure a supported consumer to process item changes asynchronously with idempotency.
Have the client block the transaction until every search-index update succeeds.
Use Route 53 health checks as the database change log.
Rely only on DynamoDB global secondary index updates and assume an index publishes full ordered change records to arbitrary external search engines automatically.

Explanation

Correct Answer

Streams captures ordered item-level changes within its retention window, allowing downstream work to decouple from the original write.

Incorrect Answers

That couples operational availability of the write path to the search service.

DNS health checks do not record item mutations.

A GSI maintains indexed table projections for DynamoDB queries; it does not by itself deliver a separate change event stream to external consumers. DynamoDB Streams captures item mutations for downstream processing.

Question 262

An IoT service emits telemetry that several independent consumers must replay for analysis. Consumers need independently tracked positions in a durable partitioned log. Which AWS option matches?

Use one SQS standard queue with competing consumers and assume every consumer gets every event.
Distribute all telemetry through one SNS-to-SQS subscription queue and assume each independent analyst replays the same full ordered history using a separate iterator.
Use a single synchronous endpoint whose unavailability blocks all producers.
Use Amazon Kinesis Data Streams with suitable shard capacity, retention and independently checkpointed consumers.

Explanation

Correct Answer

Kinesis streams retain ordered records by shard and allow separate consumer applications to read/replay within retention limits.

Incorrect Answers

A standard SQS queue distributes work between consumers rather than delivering every message to each.

One SQS queue distributes and eventually deletes messages among consumers; it does not maintain independent replay offsets for several analytics applications like a retained Kinesis stream.

That fails independent consumers and durable replay needs.

Question 263

A legacy financial application uses JMS-compatible queues with transaction semantics and cannot easily be rewritten to AWS-native SDKs. Which managed broker can preserve the protocol compatibility?

Consider Amazon MQ with a supported broker engine and compatible legacy messaging protocol.
Use only S3 object events because they natively implement arbitrary JMS queues.
Give every legacy client an IAM root credential and use SNS with no protocol adapter.
Replace the JMS queue addresses with Amazon SQS URLs without adapting JMS client libraries or validating supported transaction and acknowledgment semantics.

Explanation

Correct Answer

Amazon MQ is designed for managed brokers with familiar messaging protocols, reducing invasive changes to existing JMS/AMQP integrations.

Incorrect Answers

S3 notifications are not a general replacement for legacy broker protocols.

Root sharing does not provide JMS wire compatibility.

SQS has a native API and different delivery contracts. Simply substituting endpoint strings is not a guaranteed drop-in replacement for JMS brokers; Amazon MQ supports appropriate legacy protocols.

Question 264

A Step Functions state machine must process thousands of independent input records with bounded concurrency and aggregate their results afterward. Which native construct fits?

Manually duplicate thousands of states in one linear chain with no concurrency control.
Run one standard Parallel state with thousands of branches specified statically in the state-machine definition, assuming it creates an elastic per-item iteration automatically.
Assume a CloudFront cache automatically executes each record.
Use a Step Functions Map state with appropriate distributed/inline mode and maximum concurrency controls.

Explanation

Correct Answer

Map states orchestrate repeated item processing and can bound parallel task execution within a larger workflow.

Incorrect Answers

A serial graph adds latency and cannot express bounded fan-out efficiently.

Parallel defines known fixed branches; Map iterates over items with an explicit concurrency configuration, including Distributed Map mode for eligible large-scale workloads.

Caching does not orchestrate stateful processing jobs.

Question 265

Each customer upload to S3 should trigger asynchronous thumbnail generation after object creation, without keeping the upload HTTP request open. Which supported event integration fits?

Configure S3 object-created event notifications to a supported target such as SQS, SNS or Lambda, and make processing idempotent.
Use S3 Replication rules to copy every new object into another bucket and assume replication alone invokes the thumbnail algorithm and records an application result.
Require every uploader to wait for synchronous Lambda completion before S3 accepts a PUT.
Use S3 Glacier Deep Archive as an image processing queue.

Explanation

Correct Answer

S3 events enable asynchronous post-upload workflows, but consumers must handle possible duplicates/out-of-order notifications.

Incorrect Answers

Replication copies supported objects based on bucket rules, not arbitrary image-transformation code. S3 object-created notifications can trigger Lambda or deliver work for asynchronous processing.

Event notifications are asynchronous and do not require coupling object storage to resize completion.

Archival storage class is not a processing dispatcher.

Question 266

A research company submits thousands of independent nightly scientific jobs with varied CPU/memory needs and can tolerate queued start times. Which compute service handles managed job queues and compute environments?

Put all jobs in one always-on web server request thread.
Use S3 lifecycle transitions as executable job definitions.
Use AWS Batch with appropriate job definitions, queue priorities and EC2/Fargate compute environments.
Run AWS Batch only as an ECS service with a fixed desired count and assume an ECS service itself supplies all job-queue retries, dependencies and priorities.

Explanation

Correct Answer

AWS Batch schedules containerized batch jobs onto configured compute capacity rather than requiring a manually maintained job dispatcher.

Incorrect Answers

One synchronous web process is not a scalable batch scheduler.

Object lifecycle does not provision batch compute.

ECS manages tasks/services, but AWS Batch adds job definitions, queue priority and compute-environment scheduling for diverse batch tasks.

Question 267

A producer sends 10 MB video metadata envelopes directly as SQS message bodies. SQS rejects them for exceeding its supported maximum message size. Which pattern preserves queue-based work dispatch?

Increase the SQS message-size limit indefinitely through an IAM policy.
Store the large payload in S3 and enqueue a bounded object reference/metadata pointer for the consumer.
Split the payload into arbitrary chunks with no grouping/reconstruction metadata.
Compress and send a nominally 10 MB JSON envelope directly to SQS, assuming IAM permissions can bypass the published per-message maximum regardless of compressed size.

Explanation

Correct Answer

SQS now supports up to 1 MiB per message, not the 10 MB in this case. Store the large payload in S3, enqueue a small durable reference, grant the worker scoped GetObject access and coordinate object retention, cleanup, duplicate processing and payload integrity.

Incorrect Answers

IAM authorization cannot remove the service’s message size limit.

Ungrouped chunks can be processed incorrectly and lose data integrity.

SQS enforces its payload-size limit on actual message data. IAM cannot waive it, and compression is not a guaranteed fit for every 10 MB payload; use an S3 payload reference.

Question 268

An order-update event can be delivered twice because a producer retries after a timeout. The business cannot charge a card twice for the same order. Which processing design is necessary?

Use a stable business idempotency key, authoritative transaction-state check and duplicate-safe side-effect handling.
Assign a new payment identifier on each retry without checking previous outcomes.
Assume every SQS delivery is exactly once with no duplicates.
Use SQS message deletion as the sole evidence that payment side effects committed once, without an authoritative payment idempotency record.

Explanation

Correct Answer

At-least-once delivery is normal in distributed event systems; duplicate inputs must not create duplicate financial commits.

Incorrect Answers

A new ID may cause a second real charge.

The queue model can deliver duplicates, so business processing must be safe.

A queue acknowledgment is not a payment transaction ledger. The consumer might charge and crash before deletion; an idempotent business key protects against redelivery and retry.

Question 269

One image-enrichment consumer is unavailable for two hours, but the upload producer and unrelated consumers must continue operating. Which service design best isolates the failure?

Call every consumer synchronously and abort all uploads if one is down.
Share one in-memory list on the producer process with no persistence.
Keep one shared queue for all three consumers with a high visibility timeout, assuming the timeout creates an isolated consumer backlog and separate delivery for each team.
Buffer each consumer’s work in an independent durable queue with bounded retries and DLQs as appropriate.

Explanation

Correct Answer

Independent queues preserve producer availability and let a failed consumer recover without blocking other subscribers.

Incorrect Answers

That couples producer success to the least available downstream service.

A process crash can lose the backlog.

Visibility controls one message receipt lease, not subscriber fanout. Separate durable queues per independent team prevent the failed consumer from blocking unrelated processing.

Question 270

A task queue grows sharply during peak uploads. Workers have been scaling on CPU but many jobs are I/O-bound, so backlog increases even at low CPU. Which autoscaling metric better reflects pending work?

Scale the worker fleet exclusively on average CPU utilization and assume low CPU proves every I/O-bound request has already finished.
Use a workload-calibrated SQS backlog-per-worker or queue age target rather than CPU alone, with processing-time and scaling-lag measurements.
Disable scaling entirely because CPU is below 50%.
Use Route 53 TTL as the number of desired worker instances.

Explanation

Correct Answer

Queue backlog or age reflects delayed I/O-bound work even when CPU is low. For EC2 target tracking, raw messages visible is not necessarily proportional to worker count; normalize by active capacity or use a suitable metric and evaluate worker service-time and age limits.

Incorrect Answers

Workers can spend most of their time waiting for I/O while the queue grows. A queue age or backlog-per-worker signal can measure unfinished work more directly.

I/O-bound queues can accumulate despite low CPU.

DNS TTL is unrelated to SQS backlog or compute capacity.

Pass Your Amazon AWS Certified Solutions Architect - Associate SAA-C03 Exam Easy!

Question 260

Explanation

Correct Answer

Incorrect Answers

Question 261

Explanation

Correct Answer

Incorrect Answers

Question 262

Explanation

Correct Answer

Incorrect Answers

Question 263

Explanation

Correct Answer

Incorrect Answers

Question 264

Explanation

Correct Answer

Incorrect Answers

Question 265

Explanation

Correct Answer

Incorrect Answers

Question 266

Explanation

Correct Answer

Incorrect Answers

Question 267

Explanation

Correct Answer

Incorrect Answers

Question 268

Explanation

Correct Answer

Incorrect Answers

Question 269

Explanation

Correct Answer

Incorrect Answers

Question 270

Explanation

Correct Answer

Incorrect Answers

AWS Certified Solutions Architect - Associate SAA-C03 Practice Tests