Amazon AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 53

Question 520

An API Gateway integration invokes Lambda $LATEST, while provisioned concurrency is attached only to published alias live. API traffic still shows cold starts despite billed provisioned capacity. Which mismatch explains the observation?

Provisioned concurrency automatically applies to every invocation of $LATEST even when capacity is configured for a different published alias.
Configuring function timeout to fifteen minutes silently reattaches the provisioned warm environments to $LATEST.
Reserved concurrency is identical to provisioned initialized environments, so no alias invocation qualifier is relevant.
The API is not invoking the version/alias with provisioned environments; change the integration to use the supported qualified alias.

Explanation

Correct Answer

The paid provisioned capacity lives on alias live; API Gateway traffic to $LATEST bypasses it. Invoke the correct qualified ARN and verify the intended concurrency metrics.

Incorrect Answers

Provisioned concurrency attaches to supported published versions or aliases; calls to another qualifier are not served by that capacity.

Timeout sets an execution limit and does not redirect invocation routing to the alias with preinitialized environments.

Reserved concurrency caps simultaneous execution but does not allocate preinitialized execution environments. Qualifier routing matters.

Question 521

A Fargate Linux task is allocated 40 GiB of ephemeral storage. Its task image and uncompressed layers occupy 13 GiB, and an import extracts 29 GiB of temporary data, causing No space left on device although 29 is below 40. What explains the failure?

The 40 GiB allocation is fully available to app scratch files because image bytes are stored outside task ephemeral storage.
Increasing task CPU automatically increases ephemeral storage by exactly the same number of GiB without task-definition changes.
An IAM policy granting s3:PutObject increases free task disk even when the application is writing to local /tmp.
Pulled and expanded container images use part of Fargate task ephemeral storage, leaving less than 29 GiB for temporary files and other runtime consumption.

Explanation

Correct Answer

The combined image and extraction footprint exceeds the task disk allocation even before other runtime overhead. Increase supported ephemeral capacity or reduce/stream temporary storage.

Incorrect Answers

For supported Fargate platform versions, image layers are stored on the allocated ephemeral disk. They reduce room available for application scratch.

CPU entitlement and ephemeral disk allocation are separate Fargate task-definition properties.

S3 permissions authorize remote object operations but do not expand local disk space. The task would need to stream data or resize scratch storage.

Question 522

A distributed HPC application needs the lowest practical east-west network latency and uses an EC2 cluster placement group. Its launch of a large set of instances fails because suitable closely placed capacity is unavailable in the selected AZ. What design trade-off does this illustrate?

Cluster placement optimizes network proximity within an AZ but can constrain which physical hosts are eligible when a large coordinated fleet launches.
Cluster placement automatically spreads a single group across every AZ with identical low-latency path characteristics.
The number of snapshots in an unrelated S3 bucket directly controls eligible hardware capacity for a cluster placement group.
Spread placement and cluster placement are identical modes and both guarantee a contiguous host set for unlimited nodes.

Explanation

Correct Answer

Proximity improves latency but can constrain launch feasibility under limited colocated capacity. Plan capacity and suitable instance types/AZ availability for large clusters.

Incorrect Answers

A cluster placement group is associated with one AZ for a given group, unlike a normal regional routing policy.

S3 snapshots or bucket content are not host-topology constraints for placement groups.

Spread placement targets independent hardware for limited numbers of instances, while cluster placement targets closeness. They are distinct modes with different capacity trade-offs.

Question 523

A spread placement group already has seven running instances in one AZ. Launching an eighth in that same spread group/AZ fails even though the account has free vCPU quota. Which constraint applies?

All spread groups allow unlimited running instances per AZ if the account still has vCPU quota.
An Auto Scaling warm-up setting forces the seven-instance spread group limit, regardless of the number of racks available.
Rack-level spread placement groups allow a maximum of seven running instances per Availability Zone in a group.
Raising a CloudFront origin request quota automatically allows more spread placement group instances on the same underlying racks.

Explanation

Correct Answer

Spread placement is for a small number of critical instances on independent hardware. The documented seven-per-AZ running-instance limit explains why an eighth launch cannot fit in that group.

Incorrect Answers

Spread groups deliberately restrict colocation on underlying hardware and therefore have specific per-AZ limits independent of general account quotas.

An Auto Scaling timer changes readiness accounting, not spread placement eligibility limits.

CDN request capacity is independent of the EC2 spread-group host placement limit.

Question 524

A Cassandra cluster on EC2 uses partition placement groups with three hardware partitions. An operator places all three replicas of a key in the same partition and a single partition failure takes out every replica. What error caused correlated loss?

Partition placement groups always put every instance on exactly the same underlying hardware, even when each replica is assigned to a different partition.
The replicated dataset did not distribute replicas across independent hardware partitions; partition-aware placement must align with the replication strategy.
Choosing a Route 53 latency record for the Cassandra hostname automatically assigns database replicas to different EC2 placement partitions.
A cluster placement group guarantees independent underlying partitions for every replica while preserving Cassandra quorum after any host failure.

Explanation

Correct Answer

A partition placement group offers isolation only if replicas are intentionally mapped across different partitions; placing all copies in one partition defeats that protection.

Incorrect Answers

Different placement partitions provide hardware fault domains, so splitting replicas across them can reduce correlated risk.

DNS routing policies do not control EC2 placement-group partition assignment for individual Cassandra nodes.

Cluster placement emphasizes proximity, not partition-based replica isolation. It is not a substitute for distributing replicated state across failure partitions.

Question 525

An AWS Batch job queue contains RUNNABLE jobs, but its only connected managed compute environment has maxvCpus set to zero. Jobs are otherwise valid and there are no other scheduling restrictions. Why do they not launch?

RUNNABLE means each job has already completed and no compute environment is needed.
The attached compute environment is not allowed to provision vCPU capacity under maxvCpus=0, leaving eligible jobs waiting.
An S3 bucket prefix must match the Batch queue name for vCPUs to become available.
Adding a retry policy without changing the environment capacity provisions its first EC2 worker automatically.

Explanation

Correct Answer

With zero allowed vCPUs, a managed environment cannot scale compute to serve queued jobs. Set appropriate supported bounds and confirm allocation availability.

Incorrect Answers

RUNNABLE identifies jobs that are eligible and waiting for execution resources, not completed work.

S3 object paths do not determine Batch compute-environment vCPU limits.

Retries address job failures, not a declared maximum capacity of zero.

Question 526

An SQS job usually finishes in 20 seconds but sometimes takes nine minutes for a large video. A fixed ten-minute visibility timeout for every message delays redelivery of short failed jobs excessively. Which supported consumer-side mechanism protects long-running jobs without assigning every delivery a very long timeout?

Send DeleteMessage immediately upon receive and assume an unfinished video will still be retried after worker failure.
Reduce visibility timeout to zero after each receive and assume that hides the video until processing succeeds.
Set a suitable initial timeout and periodically call ChangeMessageVisibility as a processing heartbeat while the job is active, then delete after completion and handle duplicate deliveries idempotently.
Set SQS DelaySeconds to nine minutes on the queue and treat it as the same per-message in-flight visibility extension.

Explanation

Correct Answer

AWS supports extending per-message visibility with ChangeMessageVisibility for variable processing times, subject to a maximum visibility duration and possible duplicate deliveries.

Incorrect Answers

Deleting before work is durable can lose the job if the consumer fails; it is not a protection for long processing.

Zero visibility exposes the same in-flight message to other consumers and creates duplicate work, contrary to the requirement.

Queue delivery delay controls when new messages first become visible, not how an already received job stays hidden during processing.

Question 527

An x86_64 container includes architecture-specific native libraries. When launched unchanged on Arm-based Graviton EC2, the entrypoint fails with Exec format error. Which change is necessary?

Build and test an Arm64-compatible image/dependency set (or use a multi-architecture image manifest) before running the workload on Graviton.
Set an ALB idle timeout to 30 seconds because an HTTP timeout translates x86 instructions into Arm64 at process launch.
Choose a larger Graviton memory size because any amount of RAM permits running incompatible x86_64 native binaries natively.
Turn on an S3 Transfer Acceleration endpoint and assume downloading the image through a faster URL rebuilds its native libraries.

Explanation

Correct Answer

Different CPU architectures execute different machine-code instructions. Native extensions and image binaries need compatible Arm64 builds or a supported translation arrangement.

Incorrect Answers

A load balancer timeout controls connection behavior after services run and cannot change a binary instruction-set mismatch at startup.

Memory sizing does not translate native executable instructions. The workload must use an appropriate architecture.

Transfer speed affects image distribution, not binary compilation or CPU compatibility.

Question 528

An EC2 instance was launched with hibernation enabled, but hibernate fails and the instance shuts down normally. The 32-GiB encrypted EBS root volume contains an OS and applications taking most of the space, while the instance has 32 GiB of RAM. What prerequisite was violated?

Hibernation requires only an instance-store root volume and does not need EBS storage to hold memory.
Changing a Route 53 DNS alias allows the stop process to persist all 32 GiB of RAM even when the root disk is full.
The encrypted EBS root volume lacks enough available space for RAM contents plus OS/application data; expand a supported volume and satisfy hibernation prerequisites.
A normal EC2 stop automatically saves memory state even when hibernation requirements are not met.

Explanation

Correct Answer

The root must be encrypted, supported and spacious enough to hold the memory image as well as normal disk contents. If it cannot store the RAM state, hibernate can fail and stop becomes normal.

Incorrect Answers

EC2 hibernation stores RAM state on the encrypted EBS root volume; instance-store roots are not a replacement.

DNS naming cannot create root-volume storage capacity for a memory snapshot.

Ordinary stop preserves EBS disk but clears memory/process state. Hibernation is a specific enabled lifecycle operation with prerequisites.

Question 529

An Aurora cluster has three healthy reader instances. Reporting queries still overload the writer. Select TWO actions that directly route read-only reporting connections across the readers while retaining a safe write path.

Keep mutation transactions on the cluster writer endpoint and verify connection strings used by both application paths.
Keep every report connected to the writer but increase its backup retention period.
Connect the reporting service through the Aurora reader endpoint.
Send INSERT and UPDATE queries to the Aurora reader endpoint to balance all database traffic.
Add an RDS Multi-AZ DB-instance standby and make the report run directly against that standby.

Explanation

Correct Answer

Aurora readers are for read-only work; an explicit separation of reads and writes prevents the reporting fix from sending updates to the wrong endpoint.

The reader endpoint balances newly opened read-only connections among available reader instances; the writer endpoint does not automatically do that.

Incorrect Answers

Retention protects recovery points and cannot redirect SELECT queries off the writer CPU.

Reader endpoints do not provide write-capable reader nodes; mutations must use an appropriate writer route.

A traditional Multi-AZ DB-instance standby is for failover and is not a general query-serving endpoint; this is already an Aurora cluster with readers.

Question 530

An RDS for PostgreSQL primary serves OLTP writes and expensive nightly SELECT reports; management will accept asynchronously lagging report data but not reduced write throughput. Which design isolates the reporting queries?

Use a synchronous Multi-AZ DB-instance standby as the report destination, with the primary handling writes.
Create a suitable RDS read replica and direct the reporting connection to the replica endpoint.
Schedule the heavy SELECT reports during peak OLTP hours on the primary and raise its automatic backup retention.
Create additional RDS read replicas but keep the reporting application configured with the primary endpoint.

Explanation

Correct Answer

Read replicas offload read-only requests using asynchronous replication while the source remains authoritative for writes.

Incorrect Answers

A traditional Multi-AZ DB-instance standby is maintained for failover and is not a connection endpoint for read-only SQL.

Backup retention has no effect on query isolation; retaining reports on the primary still competes with OLTP CPU and I/O.

Adding replicas does not redirect client connections; read traffic remains on the primary until the application uses a replica endpoint.

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

Question 520

Explanation

Correct Answer

Incorrect Answers

Question 521

Explanation

Correct Answer

Incorrect Answers

Question 522

Explanation

Correct Answer

Incorrect Answers

Question 523

Explanation

Correct Answer

Incorrect Answers

Question 524

Explanation

Correct Answer

Incorrect Answers

Question 525

Explanation

Correct Answer

Incorrect Answers

Question 526

Explanation

Correct Answer

Incorrect Answers

Question 527

Explanation

Correct Answer

Incorrect Answers

Question 528

Explanation

Correct Answer

Incorrect Answers

Question 529

Explanation

Correct Answer

Incorrect Answers

Question 530

Explanation

Correct Answer

Incorrect Answers

AWS Certified Solutions Architect - Associate SAA-C03 Practice Tests