AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 49
Question 480
A securities dashboard caches current stock availability in Redis for ten hours, even though availability changes every minute. Traders see stale values after stock is exhausted. What cache policy needs correction?
- Use an additional Redis replica without changing TTL or write invalidation, expecting replicas to refresh old values automatically.
- Increase the cache TTL to thirty days so fewer backend requests occur even when stale values cause incorrect trades.
- Disable all cache invalidation events because consistent stale data is safer than frequent refreshes.
- Set suitable cache TTLs and invalidation or update-on-write strategies matched to data freshness requirements.
Explanation
Correct Answer
D:Mutable financial or inventory values require bounded staleness. Caching without timely invalidation can make the application serve incorrect operational decisions.
Incorrect Answers
A:Replication copies the cached state; it does not decide when outdated business information must be expired.
B:Longer retention magnifies data staleness rather than matching the business's minute-level change rate.
C:Avoiding invalidation preserves incorrect values beyond their validity and can violate the specified freshness requirement.
Question 481
A popular documentation site serves mostly static CSS, JavaScript and images to users on multiple continents. Every request currently travels to an S3 origin in one Region. Which cache layer reduces global latency and origin load?
- Put Amazon CloudFront in front of the origin with suitable cache policies for the static objects.
- Deploy an ElastiCache Redis instance inside the origin VPC and require global browsers to query its private cache directly.
- Increase S3 bucket object version limits and assume every client automatically receives a nearby cached copy.
- Enable RDS Multi-AZ for the documentation database even though the static site is served exclusively from S3.
Explanation
Correct Answer
A:CloudFront caches eligible HTTP objects at edge locations so users can retrieve common content closer to their geography.
Incorrect Answers
B:Private in-VPC caching is not a public global HTTP content-delivery edge network for browser asset downloads.
C:Versioning stores histories, not global content-delivery caches, so it does not reduce remote latency itself.
D:Database high availability does not provide edge caching of static S3 objects for global users.
Question 482
A financial ledger's primary write state exists only in a Redis cache with eviction enabled. A surge evicts keys and committed transactions disappear. Which architectural principle is being violated?
- Set cache TTL to zero for every key and assume this guarantees cross-region immutable storage.
- Keep committed transaction records in a durable authoritative datastore and treat cache as a replaceable acceleration layer.
- Put a CloudFront distribution before Redis and expect edge caching to become a transaction database.
- Increase Redis maxmemory and assume all committed customer transactions become permanently durable forever.
Explanation
Correct Answer
B:In-memory caches may evict data or lose state during recovery, so they should not be the sole system of record when permanent transactional correctness is mandatory.
Incorrect Answers
A:Expiration settings do not establish durability, transactional controls or immutable recovery copies.
C:Content delivery caching is not an ACID ledger datastore or a substitute for persistent transaction storage.
D:Larger capacity can delay eviction but cannot replace transactional durability and recoverable authoritative storage.
Question 483
When a user uploads a photo, four independent resizing formats must be created. The current synchronous handler performs all resizes sequentially, delaying API response by thirty seconds. Which decomposition improves latency?
- Write the source object to instance-local /tmp and assume every future worker sees the same file without a shared store.
- Keep all format conversions in the same synchronous HTTP request and increase the client timeout indefinitely.
- Queue independent resize tasks using SQS messages and let separate workers process formats in parallel.
- Use CloudFront cache invalidation as the computation engine that resizes source images into four formats.
Explanation
Correct Answer
C:Parallelizable transforms can be represented as independent jobs behind a durable buffer, allowing the upload endpoint to acknowledge quickly.
Incorrect Answers
A:Instance-local temporary storage is not shared among separately scaled workers and can vanish on replacement.
B:Serial work couples user latency to every conversion and does not take advantage of independent processing.
D:Cache invalidation removes cached content but does not render or persist transformed image variants.
Question 484
An Amazon SNS Standard topic sends notifications to an HTTPS subscription. During an extended subscriber outage the endpoint returns HTTP 503 until the configured SNS delivery retries are exhausted. The business needs undelivered notification payloads retained for investigation and controlled reprocessing, not just an alarm that a delivery failed. What should the architect configure?
- Increase the SQS message visibility timeout for a separate worker queue, assuming this controls SNS-to-HTTPS retry failures.
- Configure only an SNS delivery-failure CloudWatch alarm and assume the alert archives the failed notification bodies for later replay.
- Attach an SQS source-queue redrive policy to an unrelated SQS queue and assume SNS HTTPS subscription failures will enter that queue automatically.
- Configure the SNS HTTPS subscription redrive policy to use an authorized SQS dead-letter queue in the subscription account and Region.
Explanation
Correct Answer
D:SNS supports a subscription-level redrive policy pointing to an SQS DLQ. When SNS cannot deliver a notification to the endpoint after its retry policy, the DLQ retains the message; the queue must be in the subscription account and Region and must authorize SNS to send messages.
Incorrect Answers
A:SQS visibility timeout controls when a message already received by an SQS consumer becomes visible again. It does not retain notifications that Amazon SNS failed to deliver to an HTTPS endpoint.
B:Monitoring a delivery failure provides observability but not durable storage of an undelivered notification. A DLQ is necessary when the payload must remain available for investigation after SNS retries end.
C:SQS queue redrive policies handle consumer processing failures for messages already in a source queue. They do not replace the SNS subscription-level redrive policy used for failures delivering from SNS to an HTTPS endpoint.
Question 485
A data pipeline must run up to 400 parallel file transforms but requires visibility into each item's success and controlled concurrency. A developer proposes launching unmanaged recursive Lambda invocations from one parent handler. What orchestration alternative is more suitable?
- Run all transforms sequentially in one short Lambda execution regardless of dataset size and execution duration.
- Use only SNS delivery ordering as the map execution state machine with branch-level success history.
- Use Step Functions Map with appropriate concurrency and per-item error handling for the managed fan-out workflow.
- Allow unlimited recursive Lambda invocations without tracking any child completion or downstream capacity.
Explanation
Correct Answer
C:A Map state tracks individual branches and supports bounded parallelism, improving observability and downstream protection for bulk workflows.
Incorrect Answers
A:One serial invocation may hit duration limits and cannot meet controlled parallel throughput goals.
B:Pub-sub delivery does not implement a workflow's per-item branch state and controlled completion aggregation.
D:Unbounded child invocations can overrun quotas and obscure which file transformations failed.
Question 486
A fan-out job launches 500 API calls at once, overwhelming an RDS-backed validation service with connection errors. The orchestrator already has a concurrent worker limit setting. What should it do?
- Bound batch size and parallel invocation concurrency to the supported downstream throughput and connection capacity.
- Increase only S3 bucket replication frequency, which automatically throttles unrelated RDS API requests.
- Increase concurrency without limit because every downstream service elastically removes its own connection quotas.
- Delete all failure records so the RDS connection exhaustion metrics disappear from the monitoring dashboard.
Explanation
Correct Answer
A:Controlled fan-out prevents a durable orchestration from converting source bursts into an even larger dependency outage.
Incorrect Answers
B:S3 replication configuration does not govern a worker's database call concurrency or resource exhaustion.
C:Unbounded parallel requests intensify contention and can cause more transient failures and retries.
D:Suppressing evidence does not change database connection capacity or prevent saturation.
Question 487
An SQS dead-letter queue contains failed purchase messages. Developers propose redriving every message unchanged into the primary queue even though a parser bug remains deployed. What should be done first?
- Delete the entire DLQ without investigation to remove all visible evidence of processing failure.
- Inspect the failed payloads, correct the processing defect and selectively redrive appropriate messages.
- Disable SQS permissions on all workers permanently, because that resolves poison payload parsing.
- Redrive the complete DLQ immediately at maximum rate without examining the existing failure reason.
Explanation
Correct Answer
B:A DLQ preserves failure evidence. Redriving before correction can recreate the same poison-message loop and overload the primary consumers.
Incorrect Answers
A:Destructive cleanup loses potentially recoverable business events and prevents understanding the cause.
C:Blocking all consumers prevents work but does not fix the data or processing logic that caused the failures.
D:Unchanged bad payloads will likely fail again and can crowd out valid primary-queue processing.
Question 488
A batch Lambda reports thousands of successful invocations per hour, yet customers say some SQS orders remain unprocessed for several hours. What queue health evidence should be monitored?
- Measure only S3 bucket size even though orders are held in SQS and workers never use that bucket.
- Observe only Lambda invocation count; a high count proves every old message is finished.
- Disable queue age alarms because they cannot distinguish failed messages from completed business transactions.
- Track the oldest-message age and visible/in-flight backlog alongside processing successes and failures.
Explanation
Correct Answer
D:Successful invocation counts can coexist with a blocked subset of messages. Queue backlog and age show whether outstanding work is actually draining.
Incorrect Answers
A:Unrelated object storage metrics do not quantify queue wait time or stuck order processing.
B:An invocation can process unrelated work, fail partially or leave retries, so it is not proof that the oldest jobs progress.
C:Message age is a valuable backlog symptom when combined with consumer errors and processing throughput.
Question 489
A poison message is moved from a Standard SQS source queue to a DLQ. Investigators require ten days to examine and replay it, but the DLQ retains messages for one day. What retention strategy fits?
- Set the DLQ retention shorter than any investigation window because moving a message resets all retention needs automatically.
- Set source and DLQ retention windows to cover intended failure investigation and redrive needs, considering SQS age semantics.
- Increase only SQS visibility timeout on the source and assume the DLQ never expires messages.
- Delete the DLQ to avoid choosing any retention period for failed processing events.
Explanation
Correct Answer
B:DLQ retention must be long enough to hold messages while teams analyze and recover them. Age and expiration behavior depend on queue type and original enqueue time.
Incorrect Answers
A:A short DLQ lifetime may expire failed jobs before their root cause is corrected or redrive is approved.
C:Visibility timeout is a temporary in-flight control, not a permanent extension of DLQ message retention.
D:Without a durable failure repository, teams lose evidence and controlled replay of poison jobs.
Question 490
An ECS application runs the same image in development, testing and production, but each environment has different database endpoints and API credentials. A developer rebuilds the entire image with values hardcoded for every environment. Which configuration pattern improves control?
- Supply environment-specific settings from managed configuration and secret stores at task startup or runtime.
- Require engineers to SSH into each running Fargate task and edit configuration files manually.
- Use S3 object versioning as the sole runtime secret store without workload IAM permissions or access controls.
- Embed every environment's production password in the public image so any deployment can connect everywhere.
Explanation
Correct Answer
A:Separating deployment settings from image contents allows one validated artifact to be deployed with controlled, environment-specific parameters and credentials.
Incorrect Answers
B:Fargate tasks are ephemeral and managed; manual host editing is not a reproducible configuration deployment.
C:Object versioning is not a replacement for scoped secret retrieval and credential lifecycle management.
D:A shared image containing all secrets creates unnecessary credential exposure and violates environment isolation.
