SAP-C03 · Domain 1: Architecture Design and Implementation Built for the Cloud · 25% of the exam
Task 1.1: Design architectures built for the cloud.
Cloud-native building blocks: event-driven messaging with the right ordering and delivery guarantees, stateless services, serverless data pipelines, durable workflows, cross-account and cross-Region event routing, containers, their images and service meshes, auto scaling, ingress and edge choices, real-time WebSocket designs, and tenant isolation in multi-tenant systems.
Study it
Event-driven designs: ordering, idempotency and delivery guarantees with EventBridge, SQS, SNS and Kinesis
Try each one before opening the answer. Every option is explained, with the AWS documentation page that proves it.
Question 1 · choose 1
Twelve teams run microservices on EC2, ECS and Lambda in separate VPCs and accounts. The services call each other over HTTP, and the network team is tired of managing peering, route tables and load balancers for every new pair. Access between services must be authorized. Which approach fits?
AA Site-to-Site VPN connection between every pair of VPCs
BA full mesh of VPC peering connections with security group references
CVPC Lattice with services and client VPCs in one service network
DPublic Application Load Balancers for each service with IP allow lists
Show the answer and why
AA Site-to-Site VPN connection between every pair of VPCs
Incorrect
VPN connections link networks; they do not simplify service-to-service access or authorization.
BA full mesh of VPC peering connections with security group references
Incorrect
Peering every pair of VPCs is the routing work the team wants to stop, and peering is not transitive.
CVPC Lattice with services and client VPCs in one service network
Correct
VPC Lattice connects services across VPCs and accounts, and clients and services in the same service network can communicate if they are authorized to do so.
DPublic Application Load Balancers for each service with IP allow lists
Incorrect
Exposing internal services publicly and managing IP lists adds risk and work.
Service-to-service connectivity and authorization across VPCs and accounts is what VPC Lattice provides.
A SaaS provider is redesigning its document platform for 600 business customers. Its document storage service holds regulated data, and 40 enterprise customers require that their documents sit in resources that no other tenant uses, encrypted with their own keys. The search and notification services hold no regulated data, have no noisy-neighbor problems, and must be as cheap and simple to run as possible. Every tenant must keep the same onboarding flow, identity and operational view. Which tenant isolation design meets these requirements?
AA bridge model that silos document storage per enterprise tenant with its own KMS key and pools search and notifications
BA silo model with a separate stack of every service for each of the 600 tenants, deployed through the shared onboarding flow
CA pool model for every service, with enterprise documents kept apart by a tenant ID column that the application code filters on
DA separate installation of the product in each enterprise customer's own account, with its own onboarding, releases and operations
Show the answer and why
AA bridge model that silos document storage per enterprise tenant with its own KMS key and pools search and notifications
Correct
The bridge model mixes silo and pool per service: the regulated service gets dedicated resources for the tenants that need them, while services without those constraints stay pooled. All tenants still share one identity, onboarding and operations layer.
BA silo model with a separate stack of every service for each of the 600 tenants, deployed through the shared onboarding flow
Incorrect
Siloing every service for every tenant meets the isolation need but gives up the cost and simplicity that the search and notification services are supposed to keep.
CA pool model for every service, with enterprise documents kept apart by a tenant ID column that the application code filters on
Incorrect
In a pool model tenants share resources, so enterprise documents would not sit in resources that no other tenant uses, which the enterprise customers require.
DA separate installation of the product in each enterprise customer's own account, with its own onboarding, releases and operations
Incorrect
Separate onboarding, management and operations per customer is a managed service model, not SaaS, and breaks the requirement that every tenant keeps the same onboarding flow and operational view.
When one service has regulatory or noisy-neighbor reasons for dedicated resources and others do not, the SaaS Lens calls the mixed approach the bridge model: silo where it is required, pool everywhere else, with shared identity and onboarding across all tenants.
A team of Python developers is building an order workflow that runs almost entirely in AWS Lambda: validate the order, reserve stock, charge the card, then wait up to three days for a warehouse confirmation before shipping. The team must write the workflow as ordinary sequential Python code tested with its usual unit test framework. Completed steps must not run again after a failure, and the team does not want to pay for compute while the workflow waits. Which approach meets these requirements?
AOne Lambda function that sleeps between the steps and keeps its progress in the function's /tmp storage
BA Lambda durable function that wraps each call in a step and waits for the warehouse confirmation with a durable wait
CA Step Functions Express workflow that calls the Lambda functions and waits for the warehouse confirmation with a callback task
DA Step Functions Standard workflow written in Amazon States Language that calls a Lambda function for each step
Show the answer and why
AOne Lambda function that sleeps between the steps and keeps its progress in the function's /tmp storage
Incorrect
A Lambda invocation can run for at most 15 minutes, so it cannot wait three days, and progress kept in /tmp does not survive a new execution environment.
BA Lambda durable function that wraps each call in a step and waits for the warehouse confirmation with a durable wait
Correct
Durable functions checkpoint each step, replay without re-running completed work after an interruption, can run for up to one year, and do not incur compute charges while suspended in a wait. The workflow is plain code in a standard language.
CA Step Functions Express workflow that calls the Lambda functions and waits for the warehouse confirmation with a callback task
Incorrect
Express workflows run for at most five minutes, far short of a three-day wait.
DA Step Functions Standard workflow written in Amazon States Language that calls a Lambda function for each step
Incorrect
A Standard workflow can wait for days, but it is defined in the graph-based Amazon States Language rather than as sequential Python code, which the team requires.
For Lambda-centric workflows that the team wants to write in a standard programming language, AWS recommends Lambda durable functions: steps are checkpointed and replayed, executions can last up to a year, and waits cost nothing in compute. Step Functions fits cross-service orchestration with a visual or DSL definition.
A company will run 60 Lambda functions in one account and Region. A payment function must always be able to scale to 400 concurrent executions, even when batch functions in the same account consume all the concurrency they can get. Load tests show that total demand at peak will exceed the account's current concurrency quota. Which actions should a solutions architect take?
ARaise the memory of the payment function so that each invocation finishes sooner and needs less concurrency
BAdd a dead-letter queue to the payment function so that throttled invocations can be retried later when capacity is free
CDeploy the payment function in a second Availability Zone so that it has a separate concurrency pool
DReserve 400 concurrency for the payment function and request a higher account concurrency quota in Service Quotas
Show the answer and why
ARaise the memory of the payment function so that each invocation finishes sooner and needs less concurrency
Incorrect
More memory can shorten invocations, but it guarantees nothing when other functions take all of the account's unreserved concurrency.
BAdd a dead-letter queue to the payment function so that throttled invocations can be retried later when capacity is free
Incorrect
A dead-letter queue keeps failed asynchronous events. Payments would still be throttled at the moment they are needed.
CDeploy the payment function in a second Availability Zone so that it has a separate concurrency pool
Incorrect
Lambda concurrency quotas apply per AWS Region, not per Availability Zone, so there is no separate pool to move the function into.
DReserve 400 concurrency for the payment function and request a higher account concurrency quota in Service Quotas
Correct
Reserved concurrency sets aside that much of the account's concurrency for one function so others cannot use it, and the Regional concurrency quota can be raised through a quota increase request.
Quotas are a design input: reserve what a critical function needs, and raise the shared quota before peak demand meets it.
In a new event-driven design, a consumer service had a bug for six hours and processed events incorrectly from an EventBridge event bus. After the fix, the team wants to send those six hours of events again. What should the design include?
AA longer retry policy on the rule's target so delivery is attempted again
BCloudWatch Logs of the consumer's output
CA dead-letter queue on the rule
DAn EventBridge archive on the event bus, replayed after the fix
Show the answer and why
AA longer retry policy on the rule's target so delivery is attempted again
Incorrect
Retries apply to failed deliveries, not to events that were delivered and processed incorrectly.
BCloudWatch Logs of the consumer's output
Incorrect
Logs of output cannot resend the original events.
CA dead-letter queue on the rule
Incorrect
A dead-letter queue holds failed deliveries, not successfully delivered events.
DAn EventBridge archive on the event bus, replayed after the fix
Correct
EventBridge can archive events so you can replay them to the event bus that originally received them.
Resending past events is EventBridge archive and replay.
A new web application keeps user sessions in the memory of each web server, so users are signed out whenever an instance is replaced or scaled in. The team wants stateless web servers with fast session access. Which change fits?
AUse larger instances so they are replaced less often
BTurn on sticky sessions on the load balancer only
CStore sessions on each instance's EBS volume
DStore sessions in Amazon ElastiCache
Show the answer and why
AUse larger instances so they are replaced less often
Incorrect
Instances are still replaced and scaled in.
BTurn on sticky sessions on the load balancer only
Incorrect
Stickiness keeps a user on one server, but the session is still lost when that server goes away.
CStore sessions on each instance's EBS volume
Incorrect
Sessions stay tied to one instance.
DStore sessions in Amazon ElastiCache
Correct
ElastiCache provides caching primitives for session state, keeping it outside the web servers.
Stateless web tiers keep session state in a shared store such as ElastiCache.
A news site serves HTML pages and 2 TB of images and scripts from a fleet of EC2 web servers behind an Application Load Balancer. Readers on other continents see slow page loads, and most of the web servers' CPU goes to serving the same static files again and again. Which change improves performance the most?
AMove the web servers to the next larger instance size and raise the Auto Scaling group's maximum size
BAdd an Amazon ElastiCache cluster that the web servers use to cache the static files in memory
CPut AWS Global Accelerator in front of the load balancer to route readers over the AWS network
DServe the site through Amazon CloudFront, with the static files moved to Amazon S3 as a second origin
Show the answer and why
AMove the web servers to the next larger instance size and raise the Auto Scaling group's maximum size
Incorrect
Bigger servers add capacity but still serve every file from one Region, so distant readers stay slow and the repeated work remains.
BAdd an Amazon ElastiCache cluster that the web servers use to cache the static files in memory
Incorrect
A cache next to the servers saves some disk reads, but every byte still travels from the Region to distant readers.
CPut AWS Global Accelerator in front of the load balancer to route readers over the AWS network
Incorrect
Global Accelerator improves the network path but does not cache content, so the servers would still serve every file.
DServe the site through Amazon CloudFront, with the static files moved to Amazon S3 as a second origin
Correct
CloudFront caches content at edge locations close to readers, which cuts latency and takes repeated requests off the origin.
Repeated static content for a global audience belongs at the edge: S3 as the store and CloudFront as the cache.
A Lambda-based checkout API has a known traffic peak every evening from 6 to 9 PM, and cold starts during the peak hurt conversion. The team wants initialized environments ready for that window only. Which approach fits?
AA longer function timeout
BMoving the function into a VPC
CReserved concurrency set to the peak value during the evening
DProvisioned concurrency scheduled for the evening peak
Show the answer and why
AA longer function timeout
Incorrect
Timeouts do not prevent cold starts.
BMoving the function into a VPC
Incorrect
VPC access does not keep environments initialized.
CReserved concurrency set to the peak value during the evening
Incorrect
Reserved concurrency guarantees and caps concurrency but does not initialize environments in advance.
DProvisioned concurrency scheduled for the evening peak
Correct
Provisioned concurrency keeps pre-initialized execution environments ready to respond immediately to requests.
Predictable peaks are covered with provisioned concurrency for that time.
A company runs 40 microservices on a self-managed Kubernetes cluster in its data center and deploys them with Helm charts and kubectl. After migrating, it wants to keep the Kubernetes tooling and manifests, and it does not want to manage the control plane or choose, scale and patch worker nodes. Which container platform meets these requirements?
AAmazon ECS on AWS Fargate, with each Kubernetes manifest converted into a task definition
BA Kubernetes cluster on EC2 instances that the team installs and upgrades with an open-source tool
CAn Amazon EKS cluster with EKS Auto Mode turned on
DAmazon EKS Anywhere on the existing data center hardware
Show the answer and why
AAmazon ECS on AWS Fargate, with each Kubernetes manifest converted into a task definition
Incorrect
ECS uses its own task definitions and APIs, so the Helm charts and kubectl workflows would have to be replaced.
BA Kubernetes cluster on EC2 instances that the team installs and upgrades with an open-source tool
Incorrect
The team would keep running the control plane and the worker nodes, which it wants to stop doing.
CAn Amazon EKS cluster with EKS Auto Mode turned on
Correct
EKS runs the Kubernetes control plane, and Auto Mode extends AWS management to the infrastructure for workloads, including compute nodes, so the team keeps standard Kubernetes tooling.
DAmazon EKS Anywhere on the existing data center hardware
Incorrect
EKS Anywhere runs Kubernetes on infrastructure that the customer manages, so the company would still own the nodes.
Keep Kubernetes, drop control plane and node management: EKS with Auto Mode.
Nightly ETL scripts on a self-managed server move and reshape data from several sources into an S3 data lake, and the server fails as volumes grow. The team wants a serverless data integration service. Which service fits?
AAmazon EC2 Auto Scaling for the script server
BAWS Glue
CAmazon RDS read replicas
DAmazon CloudFront
Show the answer and why
AAmazon EC2 Auto Scaling for the script server
Incorrect
The team would still run the servers.
BAWS Glue
Correct
Glue is a serverless data integration service to discover, prepare, move and integrate data from multiple sources.
CAmazon RDS read replicas
Incorrect
Read replicas copy one database; they do not run ETL across sources.
DAmazon CloudFront
Incorrect
CloudFront delivers content; it does not transform data.