Amazon AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 31

Question 300

A Cloud Map service’s DNS record still returns an instance address that was terminated hours ago because the service instance was not deregistered and its health state is stale. What is wrong?

Service registry membership and health state were not updated when the instance was removed.
A CloudWatch alarm on S3 PUT requests directly removes Cloud Map entries.
A more expensive EC2 instance type changes a stale registry’s DNS result.
Lower Cloud Map DNS TTL to five seconds but leave the terminated instance registered as healthy in the registry, assuming TTL deletes the underlying record itself.

Explanation

Correct Answer

Cloud Map discovery requires correct registration/deregistration and supported health checks to avoid stale endpoints.

Incorrect Answers

Storage alarm metrics do not synchronize service registry membership.

Capacity alone does not update discovery records.

Short TTL reduces how long a DNS response may be cached but cannot automatically deregister a stale instance. Registration and health must reflect current running endpoints.

Question 301

A web app stores session data only on EC2 local disk and enables ALB sticky sessions. When an instance fails, users lose sessions despite automatic failover. What is the real cause?

ALB stickiness makes local disk replicate between all target instances.
Setting an EC2 Auto Scaling desired capacity of one automatically preserves all browser sessions.
Store a browser session identifier in CloudFront cache metadata and assume that ALB-target-local session files are recreated when the target fails.
Session state is tied to the failed host; sticky routing does not replicate that state to healthy hosts.

Explanation

Correct Answer

Move session state to a durable shared store or use a stateless session mechanism so target replacement is safe.

Incorrect Answers

Load balancer affinity does not implement storage replication.

Desired count does not restore host-local session bytes.

CloudFront caching cannot reconstruct the authoritative cart state stored only on the lost EC2 local disk. Shared state storage or secure stateless sessions must survive backend replacement.

Question 302

A producer publishes a shipping event successfully, but a downstream notification service is down for an hour. After recovery, the event cannot be delivered because the producer itself has no retry persistence. What architectural defect caused data loss?

The producer used AWS SDK temporary credentials, which always delete events after an hour.
Route 53 health checks must be disabled to guarantee message persistence.
Invoke the email service synchronously with an HTTP timeout and do not enqueue the successfully accepted shipping event anywhere if the service is unavailable.
The producer was coupled to a transient consumer response without storing the event in a durable queue or replayable log.

Explanation

Correct Answer

A durable SQS/event buffer allows producer acceptance while individual consumers retry their failed processing independently.

Incorrect Answers

Credential expiration affects API authentication, not durable consumer event replay.

DNS routing is not a durable event ledger.

A transient HTTP failure without a durable event buffer loses the work after the producer returns. SQS or a replayable log decouples acceptance from consumer uptime.

Question 303

An SQS-backed worker fleet scales on CPU, but remote I/O stalls each task. CPU remains at 15% while oldest-message age and visible backlog rise. What caused the scaling failure?

SQS automatically reduces the number of queue messages when CPU is below 20%.
CPU was selected as the only scaling signal even though queue pressure, not CPU saturation, limits throughput.
Increase the number of origin CloudFront cache behaviors while leaving worker scaling based only on low CPU utilization.
The queue must use a different AWS Region for every message.

Explanation

Correct Answer

Backlog-per-worker or age-of-oldest-message better captures unmet processing demand in an I/O-bound consumer.

Incorrect Answers

SQS does not clear unprocessed messages based on consumer CPU.

CDN cache configuration does not reflect SQS backlog pressure and cannot add consumers to drain an I/O-bound work queue. Scale using age/backlog indicators.

Partitioning by Region is not a prerequisite for backlog-based scaling.

Question 304

An ECS deployment updates a container’s logical port alias but clients still use an old Service Connect discovery name and fail after the rollout. What misconfiguration explains the outage?

Update the ECS task definition containerPort but retain the old Service Connect portName and clientAlias references on the deployed service.
An RDS read replica must be stopped for ECS service DNS to work.
Client endpoint names and the newly deployed ECS Service Connect service configuration disagree.
ECS services cannot perform service-to-service discovery.

Explanation

Correct Answer

Service Connect resolves logical service names configured in services; clients must use the published name and matching port.

Incorrect Answers

Service Connect references named port mappings/aliases; an inconsistent client name or port mapping can leave discovery clients using an invalid logical endpoint after deployment.

Read replica state is unrelated to Service Connect aliases.

Service Connect supports logical client endpoints when configured correctly.

Question 305

An app replaces container tasks, but clients still dial an IP address captured in an environment variable during initial deployment. Cloud Map has current healthy service instances. Why are clients failing?

Cloud Map automatically rewrites arbitrary application configuration files on every task replacement.
Create new Cloud Map service registrations correctly but keep consumers configured with a literal retired task IP instead of resolving the registered service name.
The application bypasses Cloud Map discovery and continues using a stale literal endpoint instead of re-resolving its service.
An ALB rule for a different web path is required to update the client address.

Explanation

Correct Answer

Service discovery works only when consumers query the registered service instead of hard-coding earlier task IP addresses.

Incorrect Answers

The registry exposes discovery results; it does not patch hard-coded client settings.

Registry updates are useful only to clients that query the registry. Hard-coded task addresses remain stale across ECS task replacement despite healthy registrations.

The client is not using the registry or a load balancer; the stale address is the direct cause.

Question 306

An ecommerce service moved cart sessions from EC2 process memory into a shared Redis cache, so healthy ALB targets can all read carts. But the Redis deployment has only one primary node in AZ-a and no failover replica; an AZ-a outage makes the cart service unavailable even though the EC2 tier survives in AZ-b. Which resilience control addresses the new failure boundary?

Enable ALB sticky sessions so Redis automatically replicates its only primary node to healthy AZs.
Deploy an appropriately supported Multi-AZ Redis replication group with automatic failover and client reconnection behavior, while validating session TTL and durability needs.
Move all ALB targets back into AZ-a to colocate them with the only Redis primary and eliminate multi-AZ operations.
Raise EC2 instance memory to retain each cart locally even though clients use Redis as the source of session state.

Explanation

Correct Answer

A supported Redis Multi-AZ replication group can promote a replica after an AZ outage, reducing reliance on a single cache host; applications must reconnect to the correct endpoint and tolerate appropriate transient events.

Incorrect Answers

Stickiness changes application target selection but does not create a replicated Redis primary, read replica or automatic Redis failover.

Placing all compute in the failing AZ would create an additional shared failure domain rather than preserve the multi-AZ application availability contract.

More EC2 local memory cannot repair a missing Redis replication/failover design when all application instances depend on the shared cache endpoint.

Question 307

A service must asynchronously process millions of independent thumbnail jobs and does not require ordered delivery or replay by multiple independent consumer applications. Which design avoids unnecessary streaming complexity?

Use a Kinesis Data Stream with extended record retention and let independent workers read the stream as if it were a one-consumer-per-job queue with delete acknowledgements.
Use SQS Standard with elastic consumers, idempotent handlers and an appropriate redrive policy.
Use one SQS FIFO message group for all images to force global serialization.
Keep a synchronous request open until all thumbnails complete.

Explanation

Correct Answer

A work queue is simpler for independent one-consumer-per-job dispatch; strict ordering and durable multi-subscriber log replay are not required.

Incorrect Answers

Stream retention and shard consumers suit ordered replay, but records remain accessible to consumers; they do not provide ordinary SQS job claim/deletion semantics for independent thumbnails.

Global sequencing would sacrifice throughput when jobs are independent.

This couples request latency and availability to worker duration.

Question 308

A payments processor needs order within each account but not across accounts. The workload has thousands of concurrent accounts. Which message-group design balances correctness and throughput?

Set one MessageGroupId for the entire queue.
Use SQS FIFO MessageGroupId equal to a stable account identifier, with idempotent processing.
Choose randomly changing group IDs for consecutive events from the same account.
Use SQS Standard and claim it prevents out-of-order processing of every account.

Explanation

Correct Answer

Independent accounts can advance concurrently, while one account’s records remain sequenced within its group.

Incorrect Answers

That introduces avoidable serialized processing across all accounts.

Changing groups removes the per-account ordering guarantee.

Standard queues can reorder messages and redeliver them.

Question 309

A job sometimes takes 90 seconds but typically takes 10. The team wants fast retry on failed work and no premature concurrent processing of long jobs. What visibility strategy fits?

Set the SQS visibility timeout to a fixed 10 seconds even though some legitimate jobs run 90 seconds and never extend in-flight visibility.
Set visibility to zero after receipt to maximize concurrency.
Start with a measured safe visibility timeout and extend it for active long-running jobs while retaining idempotent handlers.
Treat visibility timeout as an exactly-once completion guarantee and delete application deduplication.

Explanation

Correct Answer

Dynamic visibility extensions cover outliers while allowing recovery after genuinely failed consumers.

Incorrect Answers

Once visibility expires, other workers can receive the unfinished item concurrently. The handler should select a suitable timeout or extend it while long-running work is active.

The same in-progress job can immediately be processed by another worker.

At-least-once redelivery remains possible despite visibility settings.

Question 310

A failed-event queue holds tax filings that can be manually repaired. The team wants to avoid discarding records yet also avoid endless retries of malformed payloads. What is the best disposition?

Silently delete every message at its first transient failure.
Use a DLQ with a bounded receive threshold, traceable investigation, and controlled redrive after repair.
Retry malformed messages forever in the primary queue with no DLQ.
Configure maxReceiveCount but leave the redrive destination unset and delete poison messages after repeated receive attempts.

Explanation

Correct Answer

Poison messages should be quarantined with evidence and replayed only once their cause is addressed.

Incorrect Answers

Transient failures can be retried safely and deletion discards valuable work.

Persistent errors consume worker capacity and may delay valid work.

A bounded count alone is not enough to preserve failed filings. A valid DLQ redrive policy must retain inspectable failed work and support deliberate replay after repair.

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

Question 300

Explanation

Correct Answer

Incorrect Answers

Question 301

Explanation

Correct Answer

Incorrect Answers

Question 302

Explanation

Correct Answer

Incorrect Answers

Question 303

Explanation

Correct Answer

Incorrect Answers

Question 304

Explanation

Correct Answer

Incorrect Answers

Question 305

Explanation

Correct Answer

Incorrect Answers

Question 306

Explanation

Correct Answer

Incorrect Answers

Question 307

Explanation

Correct Answer

Incorrect Answers

Question 308

Explanation

Correct Answer

Incorrect Answers

Question 309

Explanation

Correct Answer

Incorrect Answers

Question 310

Explanation

Correct Answer

Incorrect Answers

AWS Certified Solutions Architect - Associate SAA-C03 Practice Tests