AWS Solutions Architect Associate SAA-C03 Flashcards: Architecture Decisions & Trade-Offs
Review SAA-C03 architecture choices through 300 original cards on security, resilience, performance, and cost trade-offs.
About this deck
These AWS Solutions Architect Associate flashcards give current SAA-C03 candidates 300 original prompts for choosing among plausible cloud architectures. Each answer names the decision first, then explains the constraint that makes it fit.
What the cards practice
- Requirement → architecture choice: choose a service, feature, or pattern from security, resilience, performance, and cost constraints.
- Nearby options → decisive trade-off: separate choices such as Multi-AZ from read replicas, queues from event buses, and edge caching from network acceleration.
- Failure mode → mitigation: recall the design change that limits a blast radius, preserves work, or meets a recovery target.
The deck deliberately leaves out reverse service-to-definition drills, exhaustive service inventories, copied multiple-choice questions, exam dumps, volatile scheduling and scoring details, hands-on labs, and full practice exams. Those directions either reward glossary recognition or test a different skill.
The sequence follows the current four-domain outline: 90 secure cards, 78 resilient cards, 72 high-performing cards, and 60 cost-optimized cards. It starts with identity, policy, network, and data-protection prerequisites; moves through decoupling and recovery; then applies storage, compute, database, networking, analytics, and cost trade-offs. Related variants are separated so one answer does not immediately cue the next.
An AI tool helped draft and sequence the original cards and generated the cover. Every final front and back was manually checked against the current AWS content outline and official service documentation for correctness, relevance, clarity, and one main retrieval target.
The common-knowledge service facts and AWS marks remain Amazon Web Services' property. The CC0 dedication applies only to the original wording, selection, sequence, metadata, and generated cover to the extent those elements can be dedicated. No AWS exam questions, answer keys, mark schemes, or AWS documentation text were copied; scope was checked against the official AWS Certified Solutions Architect - Associate SAA-C03 exam guide. This is an independent, unofficial study resource and is not affiliated with, endorsed by, or sponsored by Amazon Web Services.
Cards in this deck
Card 1
Question
Employees need AWS access through the company identity provider without separate IAM users. What should provide it?
Answer
AWS IAM Identity Center with federation. It gives workforce users centrally managed access and temporary credentials instead of another set of long-term identities.
Card 2
Question
An EC2 application needs S3 access without stored access keys. What identity should it use?
Answer
An IAM role attached through an instance profile. The workload receives rotating temporary credentials, so no long-term secret has to live on the instance.
Card 3
Question
A company needs a maximum-permissions guardrail across member accounts. Which control fits?
Answer
An AWS Organizations service control policy (SCP). An SCP limits what member-account principals can be allowed to do; it does not grant permissions by itself.
Card 4
Question
How should the root user of a standalone AWS account or Organizations management account be secured?
Answer
Use it only for root-required tasks, protect sign-in with MFA, and do not create root access keys. Centralized root access applies to member accounts, so these root credentials still need direct protection.
Card 5
Question
Which identities can be members of an IAM group?
Answer
IAM users only. Groups collect permissions for users; roles and other groups cannot be group members.
Card 6
Question
An operator in one AWS account needs temporary access to resources in another. What is the standard design?
Answer
Assume a cross-account IAM role. The target role trusts the source principal, and the source principal is allowed to call STS to assume it.
Card 7
Question
A developer may create roles, but those roles must never exceed a fixed permission ceiling. What should constrain them?
Answer
An IAM permissions boundary. It caps the permissions an identity-based policy can grant without supplying permissions on its own.
Card 8
Question
Which service identifies resources shared with principals outside the trusted zone?
Answer
IAM Access Analyzer. It analyzes supported resource policies to surface public and cross-account access.
Card 9
Question
A corporate directory supports SAML 2.0 and staff need console access with existing identities. What access pattern fits?
Answer
SAML federation to IAM roles. Authentication stays with the corporate identity provider while AWS issues temporary role credentials.
Card 10
Question
An S3 bucket must grant another account direct access without sharing credentials. Which policy type can do that?
Answer
A resource-based bucket policy. It names the external principal on the resource and avoids distributing the bucket owner's credentials.
Card 11
Question
Permissions must follow project ownership as resources and staff change. Which authorization model reduces policy churn?
Answer
Attribute-based access control using tags. Policy conditions compare principal and resource attributes, so matching tags drive access without one policy per project.
Card 12
Question
A company wants a governed multi-account landing zone with account provisioning and guardrails. Which service fits?
Answer
AWS Control Tower. It sets up and governs a multi-account environment on top of AWS Organizations using an opinionated landing-zone model.
Card 13
Question
An external CI system supports OIDC and needs AWS API access without stored keys. What should it use?
Answer
OIDC federation to an IAM role. Short-lived web-identity credentials remove the need to keep an AWS access key in the CI system.
Card 14
Question
What credential model distinguishes an IAM role from an IAM user?
Answer
A role supplies temporary session credentials. It is assumed when needed rather than owning a permanent password or access key.
Card 15
Question
An action is allowed by one applicable policy and explicitly denied by another. What is the result?
Answer
The request is denied. An applicable explicit deny overrides an allow during policy evaluation.
Card 16
Question
Who patches the guest operating system on an Amazon EC2 instance?
Answer
The customer. AWS secures the underlying cloud infrastructure, while the customer manages the guest OS, applications, and their configuration.
Card 17
Question
When is an IAM user more defensible than federation for human access?
Answer
Only when the access pattern cannot use roles or federation and requires a long-term identity. Treat it as an exception, restrict it tightly, and rotate credentials when required.
Card 18
Question
A third-party SaaS provider assumes customer roles. What condition helps prevent the confused-deputy problem?
Answer
A unique external ID in the role trust policy. It binds the assume-role request to the intended customer relationship.
Card 19
Question
In one account, an identity policy and a resource policy each grant different actions, with no deny. How are they evaluated?
Answer
Their allows are combined. For same-account access, either policy can contribute permission unless another applicable control restricts it.
Card 20
Question
What AWS infrastructure boundary is designed as an isolated failure domain within a Region?
Answer
An Availability Zone. Separate AZs have distinct infrastructure, so spreading components across them reduces dependence on one failure domain.
Card 21
Question
Where should an application's AWS access key not be stored?
Answer
In source code or an application image. Embedded long-term credentials leak through repositories and artifacts; use a workload role and temporary credentials.
Card 22
Question
Sensitive API calls should work only in sessions authenticated with MFA. How can this be enforced?
Answer
Use an IAM policy condition that requires MFA context. The condition narrows the allowed action to sessions that meet the stronger authentication requirement.
Card 23
Question
How can a team shrink broad permissions using observed API activity?
Answer
Generate a policy with IAM Access Analyzer, then review it. Access activity provides evidence for a narrower starting policy, while human review preserves required edge cases.
Card 24
Question
Can a service control policy grant a member-account role permission to call an API?
Answer
No. An SCP defines the outer permission boundary; an IAM or resource policy must still allow the API call.
Card 25
Question
Security operations should manage an organization-integrated service without daily use of the management account. What should be configured?
Answer
A delegated administrator account for that service. It separates operational administration from the highly privileged Organizations management account.
Card 26
Question
A role session needs fewer permissions than the role normally allows. What can narrow that session?
Answer
A session policy. It intersects with the role's permissions and cannot expand beyond what the role already allows.
Card 27
Question
How does moving from self-managed EC2 databases to a managed database change security responsibility?
Answer
AWS takes on more infrastructure and platform operations, but the customer still owns data, identities, and configuration. Managed does not mean responsibility-free.
Card 28
Question
How should an organization handle root credentials in member accounts after enabling centralized root access?
Answer
Remove the member accounts' root credentials and use centrally managed privileged root sessions when needed. Recover a member's credentials only for an exceptional task that still requires direct root sign-in, then remove them again.
Card 29
Question
Why place production and experimentation in separate AWS accounts instead of only separate VPCs?
Answer
Accounts provide a stronger isolation and policy boundary. They separate billing, quotas, identities, and blast radius while organization controls provide central governance.
Card 30
Question
Which service records account API activity for security investigation and governance?
Answer
AWS CloudTrail. Its event history and trails provide evidence of who called supported APIs, from where, and when.
Card 31
Question
A public web application must keep its EC2 application servers off the internet. What subnet layout fits?
Answer
Put the internet-facing load balancer in public subnets and the instances in private subnets. Only the load balancer accepts public traffic; security groups allow the app tier from it.
Card 32
Question
Which VPC packet filter automatically allows response traffic for an allowed connection?
Answer
A security group. It is stateful, so return traffic for an established allowed flow does not need a separate inbound or outbound rule.
Card 33
Question
A public HTTP application must block SQL injection patterns and abusive request paths. Which service is the direct fit?
Answer
AWS WAF. It filters web requests at the application layer using managed or custom rules.
Card 34
Question
Which service continuously analyzes AWS telemetry for suspicious account and workload activity?
Answer
Amazon GuardDuty. It detects threats from sources such as CloudTrail events, VPC flow data, and DNS activity without becoming an inline firewall.
Card 35
Question
IPv4 instances in a private subnet need outbound internet access but must not accept internet-initiated connections. What should the route use?
Answer
A NAT gateway in a public subnet. The private route sends outbound IPv4 traffic through it while the instances remain without direct public ingress.
Card 36
Question
Which subnet-level control requires explicit rules for both request and response traffic?
Answer
A network ACL. It is stateless and evaluates numbered allow and deny rules separately in each direction.
Card 37
Question
Which managed protection targets distributed denial-of-service attacks against supported AWS edge and load-balancing resources?
Answer
AWS Shield. Shield Standard is included for common network and transport attacks; Shield Advanced adds stronger response and protection features for higher-risk workloads.
Card 38
Question
Which service discovers and reports sensitive data in Amazon S3?
Answer
Amazon Macie. It uses managed data identifiers and classification jobs to find sensitive content and risky S3 exposure.
Card 39
Question
Private subnets need access to a supported AWS API without traversing a NAT gateway or public internet. Which construct fits?
Answer
An interface VPC endpoint powered by AWS PrivateLink. It places private endpoint network interfaces in the VPC for the service API.
Card 40
Question
An application tier should accept traffic only from the load-balancer tier, even as instance IPs change. What rule is best?
Answer
Reference the load balancer's security group in the application's inbound rule. Identity by group survives scaling and avoids brittle IP allowlists.
Card 41
Question
A public load balancer needs a trusted TLS certificate with managed renewal. Which service should issue and manage it?
Answer
AWS Certificate Manager (ACM). ACM provisions supported public certificates and handles renewal while they remain correctly associated and eligible.
Card 42
Question
Which service should assess supported compute workloads and container images for software vulnerabilities?
Answer
Amazon Inspector. It continuously scans supported EC2, container image, and Lambda package surfaces and reports prioritized findings.
Card 43
Question
Private workloads need S3 access without NAT cost or internet routing. Which endpoint type fits?
Answer
An S3 gateway endpoint. Route tables direct S3 traffic privately through the endpoint without hourly interface-endpoint charges.
Card 44
Question
A subnet needs a coarse deny rule for traffic from a known hostile CIDR. Which VPC control can express it?
Answer
A network ACL. Unlike security groups, network ACLs support explicit deny rules at the subnet boundary.
Card 45
Question
How should a global web application filter malicious HTTP requests before they reach the regional origin?
Answer
Associate AWS WAF with the CloudFront distribution. Edge filtering blocks matching requests closer to viewers and reduces unwanted origin traffic.
Card 46
Question
An application needs a database password stored securely and rotated automatically. Which service is designed for it?
Answer
AWS Secrets Manager. It stores encrypted secrets and supports managed or Lambda-based rotation workflows.
Card 47
Question
A branch office needs an encrypted connection to a VPC quickly over existing internet links. Which connection fits?
Answer
AWS Site-to-Site VPN. IPsec encrypts traffic over the internet and usually provisions faster than a dedicated private circuit.
Card 48
Question
Connections to a private instance are timing out. Which record helps determine whether network traffic was accepted or rejected?
Answer
VPC Flow Logs. They capture metadata about IP flows for investigation, though not packet payloads.
Card 49
Question
A service needs very high-performance TCP load balancing and stable IP addresses. Which load balancer fits?
Answer
A Network Load Balancer. It operates at Layer 4 and supports static IP addressing while handling connection-oriented traffic at scale.
Card 50
Question
A team needs hierarchical application configuration and encrypted parameters but no built-in database-secret rotation. Which service fits?
Answer
AWS Systems Manager Parameter Store. It suits configuration and secure strings; Secrets Manager is the stronger fit when rotation is the decisive requirement.
Card 51
Question
Does a dedicated AWS Direct Connect link encrypt traffic by itself?
Answer
No. Direct Connect supplies private dedicated connectivity, but encryption needs another mechanism such as IPsec VPN or an applicable MACsec design.
Card 52
Question
A VPC needs centrally managed stateful inspection and domain-aware egress filtering. Which service fits?
Answer
AWS Network Firewall. It provides managed network inspection with stateless and stateful rule groups for routed VPC traffic.
Card 53
Question
Where can HTTPS be terminated so private HTTP targets do not each manage a public certificate?
Answer
At an Application Load Balancer HTTPS listener. ACM integrates with the listener, centralizing public TLS while target traffic remains governed inside the VPC.
Card 54
Question
A consumer web app needs a managed sign-up directory and user authentication tokens. Which component fits?
Answer
An Amazon Cognito user pool. It manages application users and issues tokens after authentication.
Card 55
Question
An API Gateway API must be callable only from approved VPCs through private connectivity. Which API type fits?
Answer
A private API reached through an interface VPC endpoint. Its resource policy can restrict which VPC endpoints or VPCs may invoke it.
Card 56
Question
A security team must apply common WAF and network protections across many accounts. Which service centralizes those policies?
Answer
AWS Firewall Manager. It manages supported firewall and protection policies across accounts in AWS Organizations.
Card 57
Question
How should a workload obtain a secret without placing the value in an AMI or container image?
Answer
Retrieve it at runtime with a workload role. The image contains no secret, and IAM plus the secret's policy controls access to the current value.
Card 58
Question
IPv6 workloads need outbound internet access without accepting unsolicited inbound connections. Which gateway fits?
Answer
An egress-only internet gateway. It supports outbound-initiated IPv6 connectivity while blocking internet-initiated connections to the VPC.
Card 59
Question
Authenticated mobile users need temporary AWS credentials to access permitted AWS resources. Which component supplies them?
Answer
An Amazon Cognito identity pool. It exchanges supported identities for temporary role credentials with scoped AWS access.
Card 60
Question
What routing property makes a VPC subnet public?
Answer
Its route table has a route to an internet gateway. An instance still needs a public IPv4 address or IPv6 address and permissive controls to communicate directly with the internet.
Card 61
Question
S3 objects need customer-controlled key permissions and an audit trail of key use. Which server-side encryption option fits?
Answer
SSE-KMS with a customer managed KMS key. KMS adds key-policy control and CloudTrail-visible cryptographic operations beyond S3-managed keys.
Card 62
Question
Which S3 control should be the broad guardrail against accidental public buckets and objects?
Answer
S3 Block Public Access. Apply it at the account and bucket levels where public access is not required, then use explicit private policies.
Card 63
Question
Which S3 feature helps recover from accidental overwrites and deletes?
Answer
S3 Versioning. It preserves prior object versions and turns a normal delete into a delete marker instead of immediately erasing older versions.
Card 64
Question
What happens when an encrypted EBS snapshot creates a new volume?
Answer
The new volume is encrypted. EBS encryption propagates through snapshots and volumes created from those encrypted snapshots.
Card 65
Question
What is the mandatory primary authorization document for a KMS key?
Answer
The KMS key policy. Every KMS key has one, and IAM allows take effect only when the key policy enables that path.
Card 66
Question
An S3 data lake wants policy-only ownership and no object ACL management. Which setting fits?
Answer
S3 Object Ownership with Bucket owner enforced. It disables ACLs and makes the bucket owner own uploaded objects, simplifying authorization.
Card 67
Question
Compliance records must be protected from deletion or overwrite for a retention period. Which S3 feature fits?
Answer
S3 Object Lock. It provides write-once-read-many retention on versioned objects, with governance or compliance controls.
Card 68
Question
An unencrypted RDS database must become encrypted without changing it in place. What migration path fits?
Answer
Snapshot it, copy the snapshot with encryption, and restore a new database. RDS storage encryption is chosen for the new resource rather than toggled on the existing instance.
Card 69
Question
A large object must be encrypted without sending the whole payload to KMS. Which pattern fits?
Answer
Envelope encryption. A data key encrypts the payload locally, and the KMS key protects the much smaller data key.
Card 70
Question
Several applications need different S3 access policies over the same shared dataset. What reduces one giant bucket policy?
Answer
S3 Access Points. Each access point has its own endpoint and policy, letting access be separated by application or team.
Card 71
Question
What bucket prerequisite applies to S3 replication?
Answer
Versioning must be enabled on both source and destination buckets. Replication operates on object versions rather than an unversioned overwrite stream.
Card 72
Question
An application uses TLS on regional load balancers in two Regions. Where must its ACM certificates be available?
Answer
In each load balancer's Region. ACM certificates are regional resources for regional services, so each deployment needs an eligible certificate there.
Card 73
Question
What happens to existing ciphertext when a KMS key's material rotates?
Answer
It is not re-encrypted. KMS keeps the prior key material so the same logical key can decrypt older ciphertext while new encryption uses current material.
Card 74
Question
How can a bucket reject requests that do not use TLS?
Answer
Add a bucket-policy deny conditioned on insecure transport. The explicit deny blocks requests where the secure-transport context is false.
Card 75
Question
S3 data must have an automatically maintained copy in another Region. Which feature fits?
Answer
S3 Cross-Region Replication. It asynchronously copies eligible object versions to a bucket in a different Region under configured rules.
Card 76
Question
Which Region must hold an ACM certificate used by Amazon CloudFront?
Answer
US East (N. Virginia),
us-east-1. CloudFront requires its viewer certificate in that Region even when the origins live elsewhere.Card 77
Question
A key type does not support automatic KMS rotation. How can the application rotate it?
Answer
Create a replacement KMS key and repoint an alias or application configuration. Manual replacement is the rotation path for unsupported key types.
Card 78
Question
A private S3 object needs temporary download access for one user without making the bucket public. What should be issued?
Answer
A presigned URL. It delegates time-limited permission derived from the signer's credentials for a specific S3 request.
Card 79
Question
Separate teams in one Region need automatically replicated S3 copies under different accounts. Which feature fits?
Answer
S3 Same-Region Replication. It maintains eligible copies between buckets in the same Region, including cross-account destinations.
Card 80
Question
After a secret rotates, what application design prevents continued use of the old value?
Answer
Retrieve or refresh the secret at runtime instead of caching it indefinitely. Rotation only helps when consumers adopt the current version safely.
Card 81
Question
Another account must use a customer managed KMS key. What two authorization sides are required?
Answer
The key policy must trust the external account, and that account must allow its principal to use the key. Cross-account KMS access needs permission on both sides.
Card 82
Question
A CloudFront distribution should read from a private S3 origin while direct bucket access stays blocked. Which control fits?
Answer
CloudFront Origin Access Control (OAC). The bucket policy trusts signed CloudFront origin requests instead of exposing S3 publicly.
Card 83
Question
Backups must resist deletion or shortened retention, including by privileged operators. Which control fits?
Answer
AWS Backup Vault Lock in compliance mode after its grace period. It enforces write-once-read-many retention, so even privileged users cannot alter or delete protected recovery points early.
Card 84
Question
A data owner needs evidence of which S3 objects contain regulated identifiers before tightening access. What should run first?
Answer
An Amazon Macie sensitive-data discovery job. Classification findings identify likely sensitive objects so controls can follow the data's risk.
Card 85
Question
A requirement calls for customer-controlled keys in dedicated single-tenant hardware security modules. Which service fits?
Answer
AWS CloudHSM. It provides dedicated HSMs under customer control; KMS is the simpler managed choice when that control boundary is not required.
Card 86
Question
Records must be retained for a policy period and then deleted automatically. What design principle should drive storage controls?
Answer
Encode the retention policy in lifecycle and backup controls. Automated transitions, expiration, and locked retention reduce reliance on manual cleanup while matching governance requirements.
Card 87
Question
A company needs centrally governed backup schedules and retention across supported services and accounts. Which service fits?
Answer
AWS Backup with organization backup policies. It centralizes plans, vaults, monitoring, and policy-based protection for supported resources.
Card 88
Question
A shared data lake needs table-, column-, and row-level permissions across analytics services. Which service should govern it?
Answer
AWS Lake Formation. It centralizes fine-grained data-lake permissions instead of relying only on broad S3 access.
Card 89
Question
An application encrypts in one Region and must decrypt in another without a cross-Region KMS call. Which key design fits?
Answer
Related KMS multi-Region keys. Their interoperable key material supports regional cryptographic operations, while policies and grants remain regional.
Card 90
Question
How can an auditor detect whether delivered CloudTrail log files were changed or deleted?
Answer
Enable CloudTrail log file integrity validation. Digest files provide a cryptographic chain that can validate delivered log-file integrity.
Card 91
Question
A web tier must add and remove instances freely as traffic changes. What application property makes this safe?
Answer
Keep the web tier stateless. Store sessions and durable data outside individual instances so any healthy instance can serve any request.
Card 92
Question
Requests for
/ordersand/imagesmust scale as separate target groups behind one endpoint. Which component fits?Answer
An Application Load Balancer with path-based rules. Layer 7 routing sends each path to its own independently scalable target group.
Card 93
Question
A team needs a managed public API front door with authorization, throttling, and request routing. Which service fits?
Answer
Amazon API Gateway. It provides the API layer without requiring the team to run and scale gateway servers.
Card 94
Question
A producer spikes faster than workers can process jobs, but accepted work must not be lost. What should separate them?
Answer
An Amazon SQS queue. The durable buffer absorbs the spike and lets consumers drain work at their own rate.
Card 95
Question
One event must be delivered to several independent queues for parallel processing. Which messaging pattern fits?
Answer
Publish to an Amazon SNS topic subscribed by multiple SQS queues. Each consumer gets its own durable copy and failure boundary.
Card 96
Question
Events from many sources must reach different targets based on fields in each event. Which service fits?
Answer
Amazon EventBridge. Event-bus rules match event patterns and route selected events to targets without point-to-point integrations.
Card 97
Question
A business process needs branching, retries, timeouts, and visible state across several services. Which service fits?
Answer
AWS Step Functions. A state machine makes workflow control explicit instead of burying orchestration in custom glue code.
Card 98
Question
Short event-driven code must scale without server management and sits idle between requests. Which compute model fits?
Answer
AWS Lambda. It provisions execution capacity on demand and charges for requests and execution instead of continuously running servers.
Card 99
Question
A long-running container needs custom runtime dependencies, but the team does not want to manage cluster instances. Which compute option fits?
Answer
AWS Fargate with Amazon ECS or EKS. Fargate supplies task or pod compute while the team keeps the container model.
Card 100
Question
A relational database is healthy but read traffic overwhelms the writer. What should absorb the reads?
Answer
One or more read replicas. Applications route read-only queries to asynchronously replicated instances, leaving writes on the source.
Card 101
Question
Which storage type fits durable, massively scalable files accessed through APIs rather than mounted as a disk?
Answer
Object storage with Amazon S3. Objects live in buckets and are addressed by keys, which suits static assets, backups, and data lakes.
Card 102
Question
Many Linux instances across Availability Zones need a shared managed NFS file system. Which storage service fits?
Answer
Amazon EFS. It provides elastic regional file storage that multiple clients can mount concurrently.
Card 103
Question
A database on one EC2 instance needs low-latency persistent block storage. Which service fits?
Answer
Amazon EBS. It presents durable block volumes to EC2 and supports volume types tuned for different IOPS and throughput needs.
Card 104
Question
A team wants AWS-native container orchestration and has no Kubernetes requirement. Which managed orchestrator is simpler?
Answer
Amazon ECS. EKS is the fit when Kubernetes APIs and ecosystem compatibility are requirements; otherwise ECS removes that control-plane complexity.
Card 105
Question
The same database records are requested repeatedly with low-latency requirements. What layer reduces database load?
Answer
An in-memory cache such as Amazon ElastiCache. Serving hot data from memory cuts repeated database reads when the application can tolerate cache semantics.
Card 106
Question
Global users repeatedly download cacheable content from one regional origin. What should serve it closer to them?
Answer
Amazon CloudFront. Edge caching reduces origin load and viewer latency for cacheable responses.
Card 107
Question
Why split a system into independently deployable services around clear business capabilities?
Answer
To isolate change, failure, and scaling boundaries. A service can evolve or scale without forcing the whole application to move together, provided its interface stays stable.
Card 108
Question
What should exceed a queue worker's normal processing time so another worker does not immediately receive the same in-flight job?
Answer
The SQS visibility timeout. It temporarily hides a received message, but the worker must still delete it after success.
Card 109
Question
A downstream processor is slower than incoming work. Which signal should often drive worker scaling?
Answer
Queue backlog relative to processing capacity. Scaling on depth or age of work responds to actual pressure better than producer CPU alone.
Card 110
Question
A standard queue can deliver a message more than once. What property must the consumer have?
Answer
Idempotency. Reprocessing the same operation identifier should not create a second business effect.
Card 111
Question
A public API must protect a backend from sudden request floods while returning controlled errors. What should the front door enforce?
Answer
API Gateway throttling and quotas. They bound request rates before excess traffic consumes backend capacity.
Card 112
Question
Partners must upload files over SFTP while data lands in managed AWS storage. Which service avoids running SFTP servers?
Answer
AWS Transfer Family. It provides managed file-transfer protocol endpoints backed by services such as Amazon S3 or EFS.
Card 113
Question
When are containers a better fit than functions for a new service?
Answer
When the workload needs a portable custom runtime, persistent process, or execution shape that does not fit function limits. Containers trade some simplicity for more runtime control.
Card 114
Question
An asynchronous function invocation keeps failing after retries. Where should the failed event go for recovery?
Answer
An on-failure destination or dead-letter target. Preserving the event outside the function makes the failure inspectable and replayable.
Card 115
Question
Queue consumers make too many empty receive calls. Which setting reduces them?
Answer
SQS long polling. A receive request waits briefly for a message, reducing empty responses and needless API calls.
Card 116
Question
Several topic subscribers need only the events relevant to them. What prevents every subscriber from processing every message?
Answer
SNS subscription filter policies. Each subscription selects messages by attributes or message-body fields before delivery.
Card 117
Question
A team needs to reprocess past application events after deploying a corrected consumer. Which feature fits?
Answer
EventBridge archive and replay. The archive retains matching events and can resend a chosen time range to its source event bus.
Card 118
Question
A workflow can run for hours and needs durable, auditable execution with non-idempotent steps. Which workflow type fits?
Answer
A Step Functions Standard Workflow. It supports long-running stateful orchestration and an exactly-once execution model unless retries are configured.
Card 119
Question
One HTTP load balancer must route to EC2 instances, containers by IP, and Lambda targets. Which load balancer supports that model?
Answer
An Application Load Balancer. Its target groups support multiple target types while listener rules keep routing at Layer 7.
Card 120
Question
Many short-lived functions cause a relational database connection storm. What managed layer should pool connections?
Answer
Amazon RDS Proxy. It shares and reuses database connections, protecting the database from bursts of client sessions.
Card 121
Question
Existing applications use standard message-broker APIs and must decouple without a messaging rewrite. Which managed service fits?
Answer
Amazon MQ. It runs managed ActiveMQ or RabbitMQ brokers for applications that need familiar broker protocols and APIs rather than a cloud-native queue redesign.
Card 122
Question
A key-value workload needs managed horizontal scaling and single-digit-millisecond access without database servers. Which database fits?
Answer
Amazon DynamoDB. Its partitioned serverless design scales key-value and document access without instance management.
Card 123
Question
A workflow must pause until a human or external system reports completion. Which orchestration pattern fits?
Answer
A Step Functions callback with a task token. The workflow persists its state and resumes only when the external actor returns the token.
Card 124
Question
Why should presentation, application, and data tiers have separate scaling boundaries?
Answer
Each tier can match its own bottleneck and failure pattern. Scaling the whole stack together wastes capacity and increases coupling.
Card 125
Question
Jobs for each customer must stay ordered and duplicate sends should be suppressed within the queue's deduplication model. Which queue type fits?
Answer
An SQS FIFO queue with a message group per ordering boundary. FIFO preserves order within a group and supports deduplication, with less throughput flexibility than a standard queue.
Card 126
Question
Why prefer a purpose-built managed service over one generic database for every workload?
Answer
Its data model and operational behavior match the access pattern. Better fit can reduce custom code, scaling friction, and failure modes.
Card 127
Question
Traffic growth is unpredictable and downtime for resizing is unacceptable. Which scaling direction is the better default?
Answer
Horizontal scaling. Adding replaceable instances avoids one ever-larger machine and supports elastic capacity across failure domains.
Card 128
Question
A very high-volume, short, idempotent event workflow needs lower orchestration overhead. Which workflow type fits?
Answer
A Step Functions Express Workflow. It is designed for high-rate, short executions and uses at-least-once semantics for asynchronous runs.
Card 129
Question
A few malformed messages repeatedly fail and block useful retry capacity. Where should they be isolated?
Answer
In an SQS dead-letter queue. A redrive policy moves repeatedly failing messages aside for inspection instead of cycling them forever.
Card 130
Question
A regional web application must survive one Availability Zone failing. What baseline topology fits?
Answer
An Application Load Balancer and Auto Scaling group spanning at least two AZs. Healthy capacity remains reachable when one AZ is unavailable.
Card 131
Question
A relational database needs automatic failover inside one Region, not read scaling. Which deployment fits?
Answer
An RDS Multi-AZ DB instance deployment. Its synchronous standby is for high availability and cannot serve read traffic.
Card 132
Question
DNS must direct users to a standby endpoint only when the primary becomes unhealthy. Which routing policy fits?
Answer
Route 53 failover routing. Health evaluation controls an active-passive DNS response between primary and secondary records.
Card 133
Question
Which disaster-recovery pattern costs least during normal operation but usually has the longest recovery time?
Answer
Backup and restore. Only protected data and recovery artifacts are maintained until a disaster requires rebuilding the environment.
Card 134
Question
Which S3 class characteristic protects ordinary regional objects from a single Availability Zone loss?
Answer
Storage across multiple Availability Zones. Regional S3 classes redundantly store data across AZs; a one-zone class deliberately trades that resilience for cost or latency goals.
Card 135
Question
An EC2 instance fails health checks in an Auto Scaling group. What resilience behavior should follow?
Answer
Auto Scaling replaces the unhealthy instance. Desired capacity is maintained with a new instance instead of preserving a broken server.
Card 136
Question
A key-value application needs active-active writes in multiple Regions with managed replication. Which feature fits?
Answer
DynamoDB global tables. Multi-Region replicas accept local reads and writes and replicate changes between participating Regions.
Card 137
Question
Which recovery objective states the maximum acceptable time to restore service after disruption?
Answer
Recovery Time Objective (RTO). It sets the downtime target that the recovery architecture and runbook must meet.
Card 138
Question
A secondary Region should keep only core data and critical services ready, then scale the rest after a disaster. Which DR pattern fits?
Answer
Pilot light. A minimal live core shortens recovery compared with backup-and-restore while keeping steady-state cost below a warm standby.
Card 139
Question
How should a team recover consistently from a damaged server fleet without repairing instances by hand?
Answer
Replace the fleet from versioned immutable images and infrastructure definitions. Rebuilding known-good instances is more predictable than repairing drifted servers in place.
Card 140
Question
Users should resolve to the deployed Region that currently offers them the lowest network latency. Which DNS policy fits?
Answer
Route 53 latency-based routing. It selects among regional endpoints using AWS latency measurements and can combine that choice with health evaluation.
Card 141
Question
Which recovery objective states the maximum acceptable amount of data loss measured in time?
Answer
Recovery Point Objective (RPO). It determines how current replicated or backed-up data must be at recovery.
Card 142
Question
An Aurora cluster needs faster failover and read scaling. What should be added across Availability Zones?
Answer
Aurora Replicas in multiple AZs. Readers serve read traffic and provide promotion targets if the writer fails.
Card 143
Question
A smaller but fully functional copy runs in another Region and scales up during disaster. Which DR pattern is this?
Answer
Warm standby. The secondary environment already serves the complete workload at reduced capacity, trading higher steady cost for faster recovery.
Card 144
Question
What artifact lets a team rebuild network and compute resources consistently in a recovery Region?
Answer
Version-controlled infrastructure as code, such as AWS CloudFormation templates. A declared environment can be recreated and reviewed instead of reconstructed from memory.
Card 145
Question
A DNS failover record should stop returning an unhealthy external endpoint. What must inform routing?
Answer
A Route 53 health check associated with the record. DNS failover needs endpoint health evidence rather than a static primary designation.
Card 146
Question
How should critical backups be protected from compromise of the workload account?
Answer
Copy them into a separately controlled backup account. Cross-account isolation reduces the chance that one compromised administrator can destroy production and recovery copies.
Card 147
Question
Which multi-Region pattern offers the lowest failover delay but usually the highest complexity and steady cost?
Answer
Active-active. Multiple Regions serve traffic simultaneously, so recovery avoids starting a cold stack but requires distributed data and conflict handling.
Card 148
Question
A DR Region has infrastructure templates but insufficient service quota for recovery capacity. What is the design flaw?
Answer
The recovery plan cannot meet its intended scale. Quotas are regional constraints and must be raised and tested before an incident.
Card 149
Question
A service-level requirement is based on successful checkout rate, not server CPU. What should alerting monitor?
Answer
A metric tied to successful checkouts. Business-outcome signals expose availability loss that infrastructure utilization alone can miss.
Card 150
Question
A request crosses several services and intermittently slows down. Which service helps locate the slow segment?
Answer
AWS X-Ray. Distributed traces connect service calls and latency so the bottleneck can be isolated across the request path.
300 cards
AWS Solutions Architect Associate SAA-C03 Flashcards: Architecture Decisions & Trade-Offs
Study this deck for freeFlashcards opens so you can start studying.
Card 151
Question
A downstream dependency is failing and retries are exhausting every caller. Which application pattern limits the cascade?
Answer
A circuit breaker. It stops repeated calls to the unhealthy dependency for a period, preserving caller resources and allowing recovery.
Card 152
Question
A regional disaster must not remove both the primary S3 dataset and its recovery copy. Where should replication send it?
Answer
To a bucket in another Region. Cross-Region Replication separates the copy from a regional failure boundary, subject to replication timing and configuration.
Card 153
Question
An EBS recovery volume is needed in another Region. What must happen to its snapshot first?
Answer
Copy the snapshot to the target Region. EBS snapshots are regional resources even though their data is stored redundantly within the Region.
Card 154
Question
How can relational database clients reduce connection disruption during failover?
Answer
Connect through Amazon RDS Proxy. The proxy pools connections and can redirect them to the new database instance while shielding clients from some failover handling.
Card 155
Question
Which Aurora endpoint should an application use for writes so failover does not require a configured instance hostname change?
Answer
The cluster writer endpoint. It follows the current primary instance as Aurora promotes a replacement writer.
Card 156
Question
A global TCP application needs fixed anycast IP addresses and rapid routing away from unhealthy regional endpoints. Which service fits?
Answer
AWS Global Accelerator. It routes traffic over the AWS global network to healthy endpoints while keeping stable client-facing IPs.
Card 157
Question
A distribution should use a secondary origin when the primary returns configured failure responses. Which feature fits?
Answer
CloudFront origin failover with an origin group. CloudFront can retry eligible requests against the secondary origin based on selected status codes.
Card 158
Question
A global application needs one S3 access endpoint with routing across replicated buckets in multiple Regions. Which feature fits?
Answer
An S3 Multi-Region Access Point. It provides a global endpoint and can route requests among participating regional buckets.
Card 159
Question
A DynamoDB table must be restored to a recent second within its supported recovery window after accidental writes. Which feature fits?
Answer
Point-in-time recovery. It provides continuous backups and restores the table to a new table at a chosen recoverable time.
Card 160
Question
Why is a successful backup job not enough evidence for disaster readiness?
Answer
Recovery requires tested restores. A restore exercise validates permissions, dependencies, timing, data integrity, and the runbook against the stated RTO and RPO.
Card 161
Question
Why is deploying stateless compute in two Regions insufficient for regional recovery?
Answer
The application's state and dependencies also need a recovery design. Data replication, secrets, configuration, DNS, and quotas determine whether the second Region can actually serve.
Card 162
Question
How does a load balancer reduce the impact of one broken target?
Answer
It stops routing new requests to targets that fail health checks. Redundant healthy targets must still exist in other failure domains.
Card 163
Question
Many clients retry a throttled service at the same time. Which retry strategy avoids another synchronized surge?
Answer
Exponential backoff with jitter. Increasing delays reduce pressure, and randomness spreads retries instead of aligning them.
Card 164
Question
Which service assesses an application's resilience posture against defined recovery targets?
Answer
AWS Resilience Hub. It evaluates supported application resources and recommends resilience improvements and tests.
Card 165
Question
How does a durable queue keep a temporary consumer outage from becoming a producer outage?
Answer
It accepts and retains work while consumers are unavailable. Producers no longer need every consumer to respond synchronously.
Card 166
Question
What usually happens to steady-state cost as a DR design moves from backup-and-restore toward active-active?
Answer
Cost rises as recovery capacity stays more ready. Faster RTO generally requires more live infrastructure, replication, and operational complexity before a disaster.
Card 167
Question
A backup must remain recoverable after a regional disruption. Which copy strategy fits?
Answer
Maintain a backup copy in another Region. Regional separation protects recovery data from sharing the production Region's failure boundary.
Card 168
Question
A relational database needs automatic failover and readable standby capacity across three Availability Zones. Which RDS deployment fits?
Answer
An RDS Multi-AZ DB cluster. Its writer and two readable instances span three AZs, combining high availability with read capacity unlike a Multi-AZ DB instance's non-readable standby.
Card 169
Question
What should drive the first choice among object, file, and block storage?
Answer
The workload's access protocol and sharing pattern. Object storage uses API-addressed objects, file storage provides a shared hierarchy, and block storage presents volumes to compute.
Card 170
Question
A very large object must upload to S3 quickly and recover efficiently from a network interruption. Which upload method fits?
Answer
Multipart upload. Parts transfer independently and in parallel, and only failed parts need to be retried.
Card 171
Question
A general EC2 workload needs SSD block storage with performance provisioned independently of capacity. Which volume type fits?
Answer
EBS gp3. It is the general-purpose SSD choice and lets IOPS and throughput be tuned without inflating volume size solely for performance.
Card 172
Question
A shared file workload has unpredictable throughput bursts and should not manage provisioned throughput. Which EFS mode fits?
Answer
EFS Elastic Throughput. It automatically adjusts throughput to workload activity, trading manual capacity planning for usage-based scaling.
Card 173
Question
An HPC job needs a high-throughput parallel file system that can process data from S3. Which service fits?
Answer
Amazon FSx for Lustre. It provides a managed parallel file system for compute-intensive workloads and integrates with S3 datasets.
Card 174
Question
An EC2 workload needs the lowest-latency temporary scratch space and can rebuild all data after host loss. Which storage fits?
Answer
EC2 instance store. It is physically attached ephemeral block storage, so performance is high but data does not survive every stop, termination, or host failure.
Card 175
Question
On-premises file clients need low-latency cached access while durable objects live in S3. Which hybrid service fits?
Answer
Amazon S3 File Gateway. It exposes file protocols locally, caches hot data, and stores files as S3 objects.
Card 176
Question
A large active NFS dataset must move repeatedly to AWS over the network with parallel transfer and verification. Which service fits?
Answer
AWS DataSync. It accelerates and automates online data movement between supported storage systems with integrity checks.
Card 177
Question
Users far from one S3 bucket need faster long-distance uploads over the public internet. Which feature may help?
Answer
S3 Transfer Acceleration. It accepts traffic at nearby edge locations and carries it over the AWS network to the bucket; measure it because benefit depends on the path.
Card 178
Question
A latency-sensitive database needs sustained high IOPS and durability beyond general-purpose SSD. Which EBS family fits?
Answer
Provisioned IOPS SSD, typically io2. It is designed for demanding transactional workloads that need predictable provisioned I/O performance.
Card 179
Question
Windows applications need a managed shared file system with SMB and Active Directory integration. Which service fits?
Answer
Amazon FSx for Windows File Server. It supplies Windows-native file semantics and managed integration instead of adapting an NFS service.
Card 180
Question
A large sequential workload needs high-throughput HDD block storage rather than low-latency transactional IOPS. Which EBS type fits?
Answer
Throughput Optimized HDD (
st1). It targets frequently accessed, throughput-intensive sequential data such as logs and big-data workloads.Card 181
Question
Clients repeatedly need different portions of a very large S3 object. What request pattern avoids downloading it all?
Answer
Parallel or selective byte-range GET requests. Fetching only required ranges can increase aggregate throughput and reduce unnecessary transfer.
Card 182
Question
One EBS volume cannot provide enough aggregate throughput, and the application can tolerate losing the stripe if one volume fails. Which layout increases performance?
Answer
RAID 0 across multiple EBS volumes. Striping aggregates performance but adds no redundancy, so durable recovery must come from elsewhere.
Card 183
Question
A CPU-bound batch workload has little memory pressure. Which EC2 instance family category fits first?
Answer
Compute optimized. Its resource balance favors processor performance instead of paying for memory or storage the workload does not need.
Card 184
Question
An Auto Scaling group should keep average CPU near a chosen utilization as load changes. Which policy fits?
Answer
Target tracking scaling. It adjusts desired capacity to keep the selected metric near its target, like a thermostat.
Card 185
Question
A Lambda function is CPU-bound and needs to finish faster. Which configuration should be tested first?
Answer
Increase the function's memory setting. Lambda allocates more CPU with more memory, so profiling several sizes reveals the best latency and cost point.
Card 186
Question
Tightly coupled EC2 nodes need the lowest possible inter-instance network latency in one Availability Zone. Which placement strategy fits?
Answer
A cluster placement group. It packs supported instances close together for high-throughput, low-latency networking, trading away AZ-level distribution.
Card 187
Question
An in-memory analytics engine holds a very large working set in RAM. Which EC2 category fits?
Answer
Memory optimized. These instances provide a higher memory-to-compute ratio for large in-memory datasets.
Card 188
Question
One function must guarantee capacity while preventing it from consuming all account concurrency. Which setting helps?
Answer
Reserved concurrency. It guarantees up to the reserved amount for that function and also caps its maximum concurrent executions.
Card 189
Question
Thousands of queued container jobs need managed scheduling onto dynamically provisioned compute. Which service fits?
Answer
AWS Batch. It queues and schedules batch jobs while managing compute environments based on job requirements.
Card 190
Question
A group should add much more capacity for a severe alarm breach than for a small breach. Which policy fits?
Answer
Step scaling. Different breach ranges trigger different capacity adjustments, matching response size to observed pressure.
Card 191
Question
A machine-learning training job needs GPUs or purpose-built accelerators. Which EC2 category fits?
Answer
Accelerated computing instances. Their attached accelerators outperform general-purpose CPUs for supported graphics, inference, and training workloads.
Card 192
Question
A team needs managed clusters for Apache Spark and Hadoop processing over large datasets. Which service fits?
Answer
Amazon EMR. It provisions and manages big-data frameworks while retaining control over cluster sizing and engines.
Card 193
Question
A container service has no host-management requirement and each task needs a defined CPU and memory shape. Which launch model fits?
Answer
AWS Fargate. Tasks receive requested serverless container capacity without choosing or maintaining EC2 hosts.
Card 194
Question
Scale-out is slow because instances spend several minutes installing software before joining service. Which feature can reduce launch latency?
Answer
An EC2 Auto Scaling warm pool. Pre-initialized instances wait in a reusable state so they can enter service faster during scale-out.
Card 195
Question
A distributed database needs very high local I/O throughput and large instance-attached datasets. Which EC2 category fits?
Answer
Storage optimized. These instances balance compute with high sequential or random I/O for data-intensive local storage workloads.
Card 196
Question
A supported Linux workload can run on Arm and needs better price-performance. Which processor family should be benchmarked?
Answer
AWS Graviton-based instances. Native Arm-compatible applications can gain price-performance benefits, but dependencies should be verified and measured.
Card 197
Question
Users in one metropolitan area need single-digit-millisecond access to compute that still uses AWS services and APIs. Which placement may fit?
Answer
An AWS Local Zone. It places selected resources closer to that population than the parent Region, when the required services are available there.
Card 198
Question
A transactional application needs SQL joins, referential integrity, and a familiar relational engine. Which managed database family fits?
Answer
Amazon RDS or Amazon Aurora. Relational services preserve SQL and transaction semantics while AWS manages common database operations.
Card 199
Question
What key-design property prevents a high-throughput DynamoDB table from concentrating traffic on one partition?
Answer
A high-cardinality partition key with traffic spread across values. Even distribution lets DynamoDB use partitioned capacity instead of creating a hot key.
Card 200
Question
An application can tolerate stale data and wants to offload repeated relational reads. Which pattern fits?
Answer
Cache-aside with Amazon ElastiCache. The application checks the cache first, loads misses from the database, and controls invalidation.
Card 201
Question
A workload stores flexible JSON-like documents and needs MongoDB-compatible APIs on a managed service. Which database fits?
Answer
Amazon DocumentDB (with MongoDB compatibility). Its document model suits nested semi-structured records when relational joins are not the core access pattern.
Card 202
Question
A MySQL-compatible workload needs managed read scaling and fast replica promotion. Which database design fits?
Answer
Amazon Aurora with Aurora Replicas. Readers share the cluster storage layer, serve read traffic, and act as failover targets.
Card 203
Question
A DynamoDB application needs microsecond read latency for eventually consistent hot keys without application-managed cache logic. Which feature fits?
Answer
DynamoDB Accelerator (DAX). It is an API-compatible in-memory cache for DynamoDB reads.
Card 204
Question
A fraud system must traverse many relationships among accounts, devices, and transactions. Which database type fits?
Answer
A graph database such as Amazon Neptune. Graph-native relationships and traversals avoid forcing connected data into join-heavy tables.
Card 205
Question
A company needs complex SQL analytics across large structured datasets with many joins. Which service fits?
Answer
Amazon Redshift. Its columnar data-warehouse architecture is optimized for analytical queries rather than row-by-row transactional access.
Card 206
Question
Why should an application avoid routing read-after-write requests to an ordinary RDS read replica?
Answer
Replication is asynchronous, so the replica may lag. Strongly current reads should use the writer unless the engine and design provide another consistency guarantee.
Card 207
Question
A DynamoDB table needs an additional query pattern on a different partition key. What should be added?
Answer
A global secondary index. It provides another partition-and-sort-key view that can be queried without scanning the base table.
Card 208
Question
An application needs full-text search and near-real-time log analytics. Which managed engine fits?
Answer
Amazon OpenSearch Service. It indexes documents for search, aggregation, and operational analytics rather than primary relational transactions.
Card 209
Question
A microservice needs a durable primary database with in-memory access and Redis-compatible APIs. Which service fits?
Answer
Amazon MemoryDB. It combines an in-memory data model with Multi-AZ durability, unlike using an ephemeral cache as the system of record.
Card 210
Question
A bursty serverless application spends too much database capacity on opening relational connections. Which service fits?
Answer
Amazon RDS Proxy. Connection pooling and reuse let many clients share fewer database connections.
Card 211
Question
An application needs Cassandra-compatible wide-column storage without provisioning or operating database nodes. Which service fits?
Answer
Amazon Keyspaces (for Apache Cassandra). It provides managed, serverless Cassandra-compatible tables that scale without cluster administration.
Card 212
Question
Which Aurora endpoint distributes new read-only connections among available replicas?
Answer
The reader endpoint. It balances connections across Aurora Replicas; it does not split individual queries from one connection.
Card 213
Question
A load balancer must route HTTP requests by host name, path, and header. Which type fits?
Answer
An Application Load Balancer. Layer 7 listener rules understand HTTP request attributes and route them to target groups.
Card 214
Question
A website serves the same static objects to users worldwide. Which service lowers latency and origin traffic?
Answer
Amazon CloudFront. Edge locations cache objects close to viewers and fetch misses from the origin.
Card 215
Question
Dozens of VPCs and on-premises networks need transitive hub-and-spoke connectivity. Which service fits?
Answer
AWS Transit Gateway. It replaces a growing mesh of point-to-point connections with a regional routing hub.
Card 216
Question
A hybrid workload needs more consistent private network performance than internet-based VPN alone. Which connection fits?
Answer
AWS Direct Connect. A dedicated connection provides a private path with more predictable network behavior than the public internet.
Card 217
Question
A global application sends non-cacheable TCP traffic to healthy regional endpoints and needs stable anycast IPs. Which service fits?
Answer
AWS Global Accelerator. It improves the network path to regional endpoints without caching application content.
Card 218
Question
A service needs extreme TCP or UDP connection performance without HTTP-aware routing. Which load balancer fits?
Answer
A Network Load Balancer. It operates at Layer 4 and preserves a transport-level model instead of inspecting HTTP paths.
Card 219
Question
Which DNS policy chooses among regional endpoints by expected user latency?
Answer
Route 53 latency-based routing. It answers with the Region AWS expects to provide the lowest latency for the requester.
Card 220
Question
Two VPCs need direct private connectivity, with no requirement for transitive routing or a central hub. Which option fits?
Answer
VPC peering. It creates direct private routes between the two VPCs, but peering relationships are not transitive.
Card 221
Question
What placement principle reduces latency between compute and a regional data store?
Answer
Keep dependent resources in the same Region and, when the service model permits, avoid unnecessary AZ crossings. Data locality shortens the network path and can also reduce transfer cost.
Card 222
Question
A hybrid connection is needed quickly while a dedicated circuit is being provisioned. Which option provides the faster setup?
Answer
AWS Site-to-Site VPN. It uses internet connectivity and can serve as the initial or backup path while Direct Connect is arranged.
Card 223
Question
Third-party virtual firewalls must scale transparently in the network path. Which load balancer fits?
Answer
A Gateway Load Balancer. It distributes traffic across virtual network appliances while preserving an inline service-insertion model.
Card 224
Question
Should a dynamic non-HTTP application choose a CDN cache or a global network accelerator?
Answer
Choose the accelerator when traffic is non-cacheable and needs optimized routing; choose the CDN for cacheable HTTP content. The decisive constraint is caching and protocol behavior, not simply global users.
Card 225
Question
A provider must expose one service privately to many consumer VPCs without routing entire networks together. Which service fits?
Answer
AWS PrivateLink. Consumers use interface endpoints to reach the service privately without transitive VPC connectivity.
Card 226
Question
A growing multi-account network needs centrally planned, non-overlapping IP address allocation. Which service fits?
Answer
Amazon VPC IP Address Manager (IPAM). It plans, tracks, and monitors address space across accounts and Regions.
Card 227
Question
Which service moves large datasets online between on-premises storage and S3, EFS, or FSx with scheduling and verification?
Answer
AWS DataSync. It automates parallel transfer and integrity validation for supported storage endpoints.
Card 228
Question
Analysts need occasional SQL queries directly over files in S3 without running a database cluster. Which service fits?
Answer
Amazon Athena. It performs serverless SQL queries on data in place and charges by query work such as bytes scanned.
Card 229
Question
Multiple applications need to process the same ordered stream of real-time records independently. Which service fits?
Answer
Amazon Kinesis Data Streams. Consumers read retained stream records independently, with ordering maintained within each shard.
Card 230
Question
A team needs serverless data discovery, a central catalog, and managed ETL jobs. Which service fits?
Answer
AWS Glue. Crawlers and the Data Catalog describe datasets, while Glue jobs transform and load them without a persistent cluster.
Card 231
Question
An on-premises application needs iSCSI block volumes backed by cloud storage. Which hybrid service fits?
Answer
AWS Storage Gateway Volume Gateway. It presents local iSCSI volumes while integrating the data with AWS-managed cloud storage and snapshots.
Card 232
Question
Business users need managed interactive dashboards over AWS data sources. Which service fits?
Answer
Amazon Quick Sight, part of Amazon Quick. It provides managed business-intelligence analysis and published interactive dashboards rather than a data-processing engine.
Card 233
Question
Athena queries usually filter by date, but every query scans the whole S3 dataset. What layout change helps?
Answer
Partition the dataset by the common date filter and query those partitions. Partition pruning limits the files scanned for each request.
Card 234
Question
Streaming records should land in S3 or another supported destination with managed buffering and delivery. Which service fits?
Answer
Amazon Data Firehose. It buffers, optionally transforms, and delivers streaming data without consumer fleet management.
Card 235
Question
A team must upload data from its own storage devices through a high-speed AWS facility instead of ordering an AWS transfer device. Which service fits?
Answer
AWS Data Transfer Terminal. The team reserves a secure, network-ready physical location, brings its own devices, and transfers data to AWS over the facility's connection.
Card 236
Question
A data team needs Apache Spark clusters with control over versions, instances, and tuning. Which service fits better than a simple ETL job?
Answer
Amazon EMR. It provides managed big-data clusters when framework control and large-scale distributed processing are decisive.
Card 237
Question
A BI team needs repeatable complex joins across a large centralized analytical warehouse. Which service fits?
Answer
Amazon Redshift. It is built for structured warehouse analytics and high-throughput SQL reporting.
Card 238
Question
Athena queries read only a few columns from a large S3 dataset. Which file format usually reduces scanned data?
Answer
A compressed columnar format such as Parquet. Column pruning and compression avoid reading irrelevant row fields.
Card 239
Question
External partners use SFTP, but files must arrive directly in S3. Which managed endpoint fits?
Answer
AWS Transfer Family. It supports familiar managed file-transfer protocols while using AWS storage as the backend.
Card 240
Question
An application requires Apache Kafka APIs and ecosystem compatibility without operating Kafka brokers alone. Which service fits?
Answer
Amazon Managed Streaming for Apache Kafka (Amazon MSK). It preserves Kafka compatibility while AWS manages broker infrastructure tasks.
Card 241
Question
S3 access patterns are unknown and change over time, but objects need millisecond access when active. Which class fits?
Answer
S3 Intelligent-Tiering. It moves objects among access tiers based on observed use, adding a monitoring fee while avoiding retrieval fees for its automatic low-latency tiers.
Card 242
Question
A fault-tolerant batch fleet can checkpoint and lose instances with short notice. Which purchase option offers the largest discount potential?
Answer
EC2 Spot Instances. They use spare capacity at a discount, and the workload must handle interruption and capacity changes.
Card 243
Question
A new DynamoDB workload has unpredictable traffic and no reliable capacity forecast. Which mode fits first?
Answer
On-demand capacity mode. It automatically handles request volume without capacity planning, trading convenience for a higher per-request price than well-used provisioned capacity.
Card 244
Question
Private instances send heavy S3 traffic through a NAT gateway. What network change removes that NAT processing path?
Answer
Add an S3 gateway VPC endpoint and route S3 traffic through it. Gateway endpoints have no hourly or data-processing charge and keep traffic on the AWS network.
Card 245
Question
Long-lived data is rarely read, must survive an Availability Zone loss, and needs millisecond retrieval. Which S3 class fits?
Answer
S3 Standard-Infrequent Access. It stores data across multiple AZs with lower storage cost, balanced by retrieval charges and minimum-duration rules.
Card 246
Question
Compute usage is steady, but the team wants flexibility across eligible instance families, Regions, and EC2, Lambda, or Fargate. Which discount model fits?
Answer
A Compute Savings Plan. The organization commits to an hourly spend level and receives discounted eligible compute usage with broad service and configuration flexibility.
Card 247
Question
A DynamoDB workload has stable, predictable traffic that consistently uses reserved throughput. Which mode can cost less?
Answer
Provisioned capacity mode with appropriate scaling. Paying for planned read and write capacity can beat on-demand pricing when utilization is predictable.
Card 248
Question
Two chatty EC2 tiers do not need AZ separation from each other, and cross-AZ transfer dominates cost. What placement can help?
Answer
Keep the chatty path in the same Availability Zone where resilience requirements allow. Cross-AZ architecture improves failure isolation but can add data-transfer charges.
Card 249
Question
Infrequently accessed S3 data is recreatable and may be lost with one Availability Zone. Which class can cut cost?
Answer
S3 One Zone-Infrequent Access. It keeps data in one AZ, so it suits replaceable data rather than the only copy of critical records.
Card 250
Question
A stable EC2 workload needs both a billing discount and reserved capacity in one specific Availability Zone. Which Reserved Instance scope fits?
Answer
A zonal Reserved Instance. It reserves matching capacity in the chosen AZ and provides the discount there; a regional Reserved Instance provides AZ flexibility but does not reserve capacity.
Card 251
Question
A large DynamoDB table stores infrequently accessed historical data, and storage cost dominates request cost. Which table class may fit?
Answer
DynamoDB Standard-Infrequent Access. It lowers storage cost while raising request cost, so it fits storage-heavy, low-access tables.
Card 252
Question
Global users repeatedly download the same public objects from a regional origin. What can reduce origin transfer and requests?
Answer
Cache the objects with Amazon CloudFront. A high cache-hit ratio serves viewers from edge locations and reduces repeated origin delivery.
Card 253
Question
Archive data is read about quarterly but still needs millisecond retrieval. Which S3 class fits?
Answer
S3 Glacier Instant Retrieval. It lowers storage cost for rarely accessed archives while preserving real-time access, with retrieval and minimum-duration charges.
Card 254
Question
A new workload is short-lived, cannot be interrupted, and has unknown steady demand. Which EC2 purchase option avoids commitment?
Answer
On-Demand Instances. They provide full flexibility without a term commitment, making them suitable while demand is uncertain.
Card 255
Question
A relational workload is intermittent and must scale database capacity without manual instance changes. Which option fits?
Answer
Aurora Serverless v2. It adjusts Aurora capacity in fine-grained steps, which suits variable demand that would leave provisioned instances idle.
Card 256
Question
What is the main trade-off between one shared NAT gateway and one NAT gateway per Availability Zone?
Answer
A shared gateway can cost less, while per-AZ gateways improve fault isolation and avoid cross-AZ routing for local private subnets. Choose from availability and traffic economics together.
Card 257
Question
Archives are read roughly yearly and can wait minutes or hours for restore. Which S3 class fits?
Answer
S3 Glacier Flexible Retrieval. It trades immediate access for lower archive storage cost and several retrieval-speed options.
Card 258
Question
An EC2 fleet averages low CPU and memory use for months. What should happen before buying a larger commitment?
Answer
Right-size the instances using measured utilization. Commitments discount usage, but they do not fix paying for capacity the workload never needs.
Card 259
Question
An RDS database has a stable multi-year production baseline. Which purchase choice can discount its instance cost?
Answer
Reserved DB instances. A term commitment discounts predictable RDS instance usage, after the database has been right-sized.
Card 260
Question
When can an interface VPC endpoint cost more than keeping low-volume service traffic on an existing NAT path?
Answer
When endpoint hourly and data-processing charges exceed the avoided NAT and transfer cost. Private routing is valuable, but endpoint count and traffic volume still matter.
Card 261
Question
Compliance archives are almost never read and can tolerate the longest restore time. Which S3 class minimizes storage cost?
Answer
S3 Glacier Deep Archive. It is the lowest-cost S3 archive tier for very rare access, with long minimum duration and hours-scale retrieval.
Card 262
Question
Traffic varies through the day and idle EC2 instances dominate cost. Which capacity strategy fits?
Answer
Auto Scaling with sensible minimum and maximum capacity. Matching instance count to demand removes idle capacity while preserving required headroom.
Card 263
Question
An RDS instance has consistently low CPU, memory, and I/O demand. What should reduce cost before changing engines?
Answer
Right-size the DB instance and storage to measured needs. A smaller supported configuration removes idle capacity with less migration risk than an engine redesign.
Card 264
Question
When can a dedicated hybrid link become economical compared with internet VPN?
Answer
When sustained data volume, required consistency, or private connectivity justifies its port and provider costs. VPN usually has a lower entry cost for modest or temporary traffic.
Card 265
Question
Objects become cold predictably after 30 days and expire after a policy period. What should automate the change?
Answer
An S3 Lifecycle rule. It transitions eligible objects to cheaper classes and expires them on a schedule without manual jobs.
Card 266
Question
A small event handler runs for seconds a few times per hour. Which compute model avoids paying for an idle server?
Answer
AWS Lambda. Request-based execution suits intermittent work when the function limits and event model fit.
Card 267
Question
When is a cache more cost-effective than adding another relational read replica?
Answer
When many requests reuse the same hot data and can tolerate cache freshness rules. A cache can remove repeated reads; a replica is better for broad SQL read scaling.
Card 268
Question
A small branch sends modest traffic to AWS and does not need a dedicated circuit. Which connection is the lower-entry-cost fit?
Answer
AWS Site-to-Site VPN. It uses the internet and avoids dedicated-link provisioning, with less predictable performance than Direct Connect.
Card 269
Question
A general SSD volume is overprovisioned only because its older type ties performance to size. Which EBS type can decouple them?
Answer
EBS gp3. Capacity, IOPS, and throughput can be tuned independently, which avoids buying unused storage just to reach performance targets.
Card 270
Question
A small container workload has irregular demand and the team cannot keep EC2 hosts well utilized. Which launch model may cost less operationally?
Answer
AWS Fargate. Per-task capacity avoids an idle host fleet, though steady high utilization should be compared with optimized EC2 capacity.
Card 271
Question
Automated database backups are kept far longer than the business recovery policy requires. What should change?
Answer
Align backup retention with the validated recovery requirement. Extra retention adds storage cost without value when legal, audit, and recovery needs do not require it.
Card 272
Question
When does Transit Gateway usually justify its attachment and processing charges over many peerings?
Answer
When centralized, transitive connectivity replaces a complex VPC and hybrid mesh. The operational simplicity can outweigh service charges at larger network scale.
Card 273
Question
Which EBS type is cost-effective for large, frequently accessed sequential workloads that need throughput rather than transactional IOPS?
Answer
Throughput Optimized HDD (
st1). It fits big sequential scans such as logs or data processing, not boot volumes or small random I/O.Card 274
Question
A Linux application and its dependencies support Arm. What compute change should be benchmarked for price-performance?
Answer
Move to Graviton-based instances or runtimes. AWS-designed Arm processors can improve price-performance when the software stack is compatible.
Card 275
Question
A relational database is write-bound, while read replicas sit mostly idle. What does that imply?
Answer
Read replicas do not address the actual bottleneck and may be removable. They scale reads; write capacity needs another design or database choice.
Card 276
Question
Three VPCs need a few direct, non-transitive private links. Which connectivity can avoid a paid routing hub?
Answer
VPC peering. A small set of direct relationships can be simpler and cheaper, but the design becomes hard to manage as the mesh grows.
Card 277
Question
Most files in an EFS file system go untouched after a few weeks. What should reduce storage cost automatically?
Answer
EFS lifecycle management. It moves inactive files to lower-cost EFS storage classes according to the configured policy.
Card 278
Question
Capacity demand rises at the same known time every weekday. Which scaling method can add capacity before the rise?
Answer
Scheduled scaling. A recurring action changes desired or boundary capacity at the known time instead of waiting for utilization to climb.
Card 279
Question
Which managed service moves database data with ongoing replication to reduce migration downtime?
Answer
AWS Database Migration Service (AWS DMS). It copies data and can keep source and target synchronized during a migration.
Card 280
Question
Private workloads reach DynamoDB through a NAT gateway. Which endpoint can remove that path without hourly endpoint charges?
Answer
A DynamoDB gateway VPC endpoint. Route-table entries keep supported traffic private and avoid NAT processing for that service.
Card 281
Question
Old EBS snapshots accumulate after every backup cycle. What policy prevents indefinite spend?
Answer
An automated retention policy that expires unneeded snapshots. Recovery requirements should set how many copies and ages remain, rather than keeping every snapshot.
Card 282
Question
Demand follows a recurring pattern but its required capacity changes over time. Which scaling method forecasts it?
Answer
Predictive scaling. It analyzes historical load and schedules capacity ahead of forecast demand, while dynamic scaling can handle deviations.
Card 283
Question
A database migration changes from one engine family to another and the schema is incompatible. What tool handles much of the conversion?
Answer
AWS DMS Schema Conversion. It converts schema objects and much database code for a heterogeneous migration; AWS DMS then moves and replicates the data.
Card 284
Question
What is the first architecture question when network transfer becomes a major AWS bill item?
Answer
Why is the data crossing a Region, Availability Zone, or internet boundary? Removing unnecessary movement often beats negotiating around the same inefficient path.
Card 285
Question
A data publisher wants approved external users to pay request and download charges for a shared S3 dataset. Which feature fits?
Answer
S3 Requester Pays. Authenticated requesters identify themselves as the payer, while the bucket owner still pays storage charges.
Card 286
Question
Development instances sit unused every night and weekend. What is the simplest cost control?
Answer
Stop or scale them down on a schedule. Non-production availability requirements rarely justify paying for full capacity when nobody uses it.
Card 287
Question
Expired session records should leave a DynamoDB table automatically without a cleanup fleet. Which feature fits?
Answer
DynamoDB Time to Live (TTL). An expiry attribute marks items for background deletion without consuming application write capacity for each cleanup.
Card 288
Question
Millions of tiny payloads create high request overhead and poor network efficiency. What transport change can help?
Answer
Batch and compress data where latency requirements allow. Fewer larger transfers reduce per-request overhead and bytes moved.
Card 289
Question
What decides between DataSync and Data Transfer Terminal for a large migration using customer-owned storage devices?
Answer
Whether the source network can meet the deadline and cost target. Use DataSync remotely over a suitable network; use Data Transfer Terminal when bringing the devices to a high-speed AWS facility is the better transfer path.
Card 290
Question
How should a Spot fleet reduce the chance that one capacity pool interrupts all workers?
Answer
Diversify across instance types and Availability Zones. More eligible pools improve the fleet's chance of finding spare capacity and surviving a constrained pool.
Card 291
Question
How can choosing a purpose-built database reduce total cost even when its unit price is not the lowest?
Answer
It removes mismatched infrastructure and custom workaround code. The right data model can need fewer resources and less operations work for the required access pattern.
Card 292
Question
A metered downstream API can process a fixed rate, but clients create expensive bursts. What control protects cost and capacity?
Answer
Throttle requests and buffer acceptable asynchronous work. A rate boundary prevents uncontrolled consumption while a queue preserves work that need not finish immediately.
Card 293
Question
A provisioned file system has far more capacity and throughput than monitoring shows it uses. What is the first cost action?
Answer
Right-size it against measured demand and headroom. Paying for unused provisioned capacity is avoidable when a smaller setting still meets performance and growth requirements.
Card 294
Question
Which service recommends rightsizing for supported compute resources from utilization data?
Answer
AWS Compute Optimizer. It analyzes metrics and recommends configurations, which should still be checked against workload constraints.
Card 295
Question
An RDS database grows unpredictably, and permanently overallocating storage wastes money. Which feature helps?
Answer
RDS storage autoscaling. It increases allocated storage when needed within a configured maximum, reducing the need for a large speculative allocation.
Card 296
Question
A dedicated connection is provisioned far above observed peak throughput. What should happen before renewal?
Answer
Right-size bandwidth against measured demand and required headroom. Excess fixed capacity adds cost without improving an unconstrained workload.
Card 297
Question
Which service feature gives organization-wide S3 usage and activity metrics to find cost and protection opportunities?
Answer
S3 Storage Lens. Its aggregated metrics and recommendations expose storage patterns that can guide lifecycle, cleanup, and protection changes.
Card 298
Question
A large in-memory application needs long idle periods but expensive startup from scratch. Which EC2 state may balance cost and restart time?
Answer
EC2 hibernation. It saves RAM state to the encrypted root EBS volume and stops instance compute billing, subject to hibernation requirements and storage cost.
Card 299
Question
What evidence should drive a database cost change: list price alone or workload metrics plus access patterns?
Answer
Workload metrics plus access patterns. Capacity, consistency, latency, availability, and operational needs determine whether a cheaper-looking engine is actually cheaper to run.
Card 300
Question
How can a company identify which applications drive shared network and transfer spend?
Answer
Use consistent cost allocation tags and detailed billing data. Ownership metadata lets Cost Explorer or the Cost and Usage Report attribute spend before teams optimize it.
300 cards
AWS Solutions Architect Associate SAA-C03 Flashcards: Architecture Decisions & Trade-Offs
Flashcards opens so you can start studying.