AWS Certified Solutions Architect - Associate SAA-C03 Practice Test 72
Question 710
A tightly coupled simulation needs frequent low-latency messages among EC2 worker nodes. The team has measured inter-node networking as its main bottleneck and can accept a correlated failure domain. Which placement feature merits evaluation?
- Use a spread placement group because spreading nodes across racks always minimizes round-trip latency between each pair.
- Use an appropriate EC2 cluster placement group with compatible instance types and capacity planning.
- Use a partition placement group solely to force every worker onto one rack for minimum latency.
- Disable enhanced networking on the nodes to improve the exchange rate.
Explanation
Correct Answer
B:Cluster placement groups can improve inter-instance network proximity for latency-sensitive/high-throughput computation, at the cost of placement constraints.
Incorrect Answers
A:Spread placement primarily targets hardware isolation rather than tightly clustered network performance.
C:Partition groups separate partitions for failure isolation, not a guarantee of optimal all-to-all latency.
D:Network acceleration features generally help rather than hinder compatible high-rate messaging.
Question 711
A small fleet includes independent security-monitoring nodes. A hardware failure must not terminate every node at once; there is no performance requirement for all-to-all low-latency compute. Which placement approach suits?
- Place every independent monitor in a cluster placement group solely to maximize hardware isolation.
- Run all monitors as containers on one EC2 host because a single host is equivalent to spread placement.
- Use a spread placement strategy within supported limits, or appropriately separated AZs, to reduce common hardware failure exposure.
- Use one EBS Multi-Attach volume as a substitute for spreading the monitor compute.
Explanation
Correct Answer
C:Spread placement emphasizes distinct underlying hardware rather than collocating tightly coupled instances.
Incorrect Answers
A:Cluster placement groups favor proximity and do not maximize physical failure-domain separation.
B:Host consolidation makes one hardware loss remove all monitors.
D:Shared block storage does not create independently surviving monitor instances.
Question 712
An HPC application constantly exchanges large state vectors across AZs. The team sees both elevated inter-AZ network latency and transfer charges. It has no specific cross-AZ fault-tolerance requirement for this transient stage. What should be measured before redesign?
- Place every communicating worker in a different Region and assume this lowers inter-node transfer delay.
- Scale only EBS IOPS because all cross-AZ network delay is actually disk latency.
- Disable application checkpointing and ignore that AZ placement can concentrate failure exposure.
- Benchmark placing the tightly coupled compute nodes within an appropriate AZ/placement group while persisting results redundantly elsewhere.
Explanation
Correct Answer
D:Compute locality can reduce transfer latency/cost for chatty HPC stages; durability and fault tolerance can be handled separately for outputs.
Incorrect Answers
A:Greater geographic separation typically increases latency and adds transfer-cost categories.
B:Disk I/O tuning does not remove network distance or billed inter-AZ movement.
C:Performance optimization should account for the recoverability tradeoff of any colocation.
Question 713
An ECS service behind an ALB deploys new tasks in rolling batches. Some requests take two minutes to finish, but the application stops old tasks almost immediately after deregistration, causing intermittent 5xx responses on those long requests. Which load-balancer attribute should be considered along with the application shutdown behavior?
- Configure an appropriate target-group deregistration delay and allow draining tasks to finish in-flight work before they exit.
- Reduce the ALB target-group deregistration delay to a few seconds and increase the Route 53 TTL, assuming old in-flight requests survive the stopped task.
- Disable all ALB target health checks permanently to avoid task replacement.
- Change only the ECS task name; the ALB then preserves every live TCP connection indefinitely.
Explanation
Correct Answer
A:ALB stops sending new requests to a deregistering target and supports a configurable delay while in-flight requests finish. The task termination grace and application shutdown path must also accommodate the drain period.
Incorrect Answers
B:A short deregistration delay and terminated backend can interrupt slow in-flight requests. DNS caching operates at a different layer and cannot preserve a terminated target’s active HTTP response.
C:Health checks are needed to detect failing targets and are not a substitute for graceful shutdown.
D:Renaming a task does not create connection-draining semantics and cannot guarantee unbounded connection lifetimes.
Question 714
A globally accessed financial application already uses a standard AWS Global Accelerator with two healthy regional endpoints. A stateful TCP client opens several short connections with different source ports. Under the default listener configuration, these connections can reach different endpoints even when they originate from one public client IP. The application team needs endpoint affinity by client IP, while accepting that failures can still cause rerouting. What should be changed?
- Leave listener affinity set to NONE and raise the endpoint health-check interval; flow hashing will then ignore source ports.
- Configure SOURCE_IP client affinity on the Global Accelerator listener so connections sharing a source IP are preferentially assigned to the same endpoint.
- Enable an ALB cookie-based sticky session on an unrelated application because that makes Global Accelerator choose one endpoint for every TCP protocol.
- Set the route control in Route 53 to a single geolocation answer and expect it to override Global Accelerator listener connection hashing.
Explanation
Correct Answer
B:SOURCE_IP affinity uses source and destination IP values for endpoint selection, so connections from a given source IP tend to use the same endpoint. Regional disruption can override affinity, so application state still needs resilience.
Incorrect Answers
A:With NONE affinity the standard listener hashes the five-tuple, including source port; changing health checks does not remove that variation from connection endpoint selection.
C:An application load balancer cookie may influence HTTP target selection for clients that honor cookies, not the Global Accelerator listener’s TCP connection-affinity algorithm across regional endpoints.
D:Route 53 DNS policies are separate from the transport listener hashing used by an existing Global Accelerator. Replacing the DNS answer does not configure source-IP client affinity within the accelerator.
Question 715
A security architecture requires transparent insertion of third-party packet-inspection appliances into VPC network paths, with scaled appliance targets. Which balancing service is purpose-built for this?
- Use ALB path-based routing to forward all raw non-HTTP packets into an intrusion prevention appliance.
- Use an S3 gateway endpoint as a general-purpose firewall packet interception device.
- Use Gateway Load Balancer with compatible appliance targets and endpoint-based routing architecture.
- Use Network Load Balancer alone without appliance insertion routes and assume VPC traffic automatically passes through it.
Explanation
Correct Answer
C:GWLB is designed to distribute inline network traffic to virtual appliances while preserving the required traffic flow patterns.
Incorrect Answers
A:ALB is an HTTP application routing component and cannot transparently inspect arbitrary IP packets.
B:An S3 gateway endpoint is a service-routing mechanism limited to eligible object API traffic.
D:NLB can distribute traffic to registered targets but is not automatically inserted into every VPC flow.
Question 716
An ALB has listeners and matching target groups, but every target fails health checks and newly launched Auto Scaling instances never register. Engineers claim choosing ALB already guarantees application capacity. What is the missing operating responsibility?
- Increase the Auto Scaling group desired count while leaving the launch template application health-check port wrong.
- Create an additional ALB listener that forwards to the same empty or unhealthy target group.
- Change the Route 53 alias to point to a new ALB using the same failing target registration and health-check behavior.
- Configure target registration, health checks and enough healthy backend capacity, with Auto Scaling integration as needed.
Explanation
Correct Answer
D:An ALB requires registered targets that pass application-relevant health checks. Connect the Auto Scaling group to the intended target group, confirm listener, port, security-group and readiness settings, and maintain enough healthy workers across failure domains.
Incorrect Answers
A:Launching more instances does not make them healthy if registration and readiness checks remain misconfigured.
B:Listener configuration changes routing entry points, not target-registration state or application availability.
C:Replacing the DNS target does not provide healthy backend compute; requests still fail if no target is properly registered and ready.
Question 717
An Aurora PostgreSQL cluster has one writer and three Aurora reader instances. The application already provisions the reader capacity but sends every reporting connection to the cluster writer endpoint. Engineers want client connections for SELECT-only traffic distributed across readers without hardcoding individual instance names. What address should the client use?
- Connect the reporting clients to the Aurora cluster reader endpoint, with separate writer-endpoint connections for write transactions.
- Continue sending all SELECT-only connections to the writer endpoint and assume it automatically distributes them to reader instances.
- Send UPDATE statements to the reader endpoint because it guarantees read/write balancing over the three readers.
- Use the RDS API endpoint as the SQL query target, which serves as a direct connection-balancing read replica.
Explanation
Correct Answer
A:The Aurora reader endpoint balances newly established read-only connections among available Aurora Replicas, reducing the read demand on the writer. It balances connections, not each individual SQL query.
Incorrect Answers
B:The cluster writer endpoint reaches the primary writer, so sending all reporting there bypasses the existing reader capacity.
C:Reader replicas do not provide the normal writable primary endpoint for application transactions.
D:RDS control-plane APIs are not a substitute for an Aurora database endpoint used by clients.
Question 718
An RDS database has three read replicas, but CPU on the primary remains saturated. Application configuration shows every SELECT still uses the primary writer endpoint. What change is required?
- Create additional replicas without changing the connection strings used by read-only queries.
- Route suitable read-only sessions through the replica instance endpoints or a deliberate application read-routing layer.
- Enable a database performance monitoring feature on the writer and expect its observations to route application SELECT statements to replicas.
- Increase primary RDS storage throughput while leaving every read-only session connected to the writer despite the existing idle replicas.
Explanation
Correct Answer
B:Read replicas do not automatically absorb application traffic; clients must target their endpoints and account for read freshness.
Incorrect Answers
A:Unused replicas add cost while the original primary handles the same SQL workload.
C:Performance monitoring can reveal expensive read queries, but cannot change clients that still connect to the writer endpoint. The application needs explicit read-only routing or replica-aware connection logic.
D:Faster storage may help some I/O-bound reads, but it does not distribute CPU-heavy SELECT execution to the existing replica instances. The scenario identifies endpoint selection as the missing step.
Question 719
An account-balance page must show a payment immediately after commit. Read replicas lag the writer by several seconds during bursts. Which routing principle preserves read-after-write expectations?
- Route the immediate confirmation read to any asynchronous replica without checking lag.
- Increase the number of read replicas and assume every replica becomes synchronously current.
- Read the authoritative writer or implement an application consistency strategy for the immediate post-commit read.
- Set the user's browser cache TTL to one day to guarantee a fresh post-payment balance.
Explanation
Correct Answer
C:Asynchronous replicas can return pre-commit snapshots; strict read-after-write operations may need primary reads or explicit freshness handling.
Incorrect Answers
A:A nonzero replication delay can display an outdated account balance.
B:Replica quantity does not eliminate asynchronous propagation semantics.
D:Longer browser caching works against immediate data freshness.
Question 720
A slow RDS report uses a full scan and a nested loop on millions of rows because a filter column has no useful index. The team proposes adding five read replicas as the first optimization. What is better?
- Add replicas and force the query to run identically on each without changing the expensive full scan.
- Increase CloudWatch log retention and assume RDS uses it as a query index.
- Replace the database's security group with a wider one to make joins avoid table scans.
- Inspect the query plan and create or tune appropriate indexes and SQL access patterns before purchasing replica capacity.
Explanation
Correct Answer
D:An inefficient query plan can remain expensive on every replica; improving data access often reduces work more directly.
Incorrect Answers
A:More hosts can spread calls, but individual queries will still perform wasteful reads.
B:Log retention does not improve relational query access paths.
C:Network authorization is unrelated to a missing relational index.
