SAA-C03 question patterns

Lambda, Fargate, Aurora, DynamoDB, S3 or another managed service instead of instances you patch and scale.

Anything with servers, scripts or manual steps adds operational work.

Commit to a Compute Savings Plan or Reserved Instances for the baseline.

On-Demand is the most expensive way to run something 24/7.

Run the fault-tolerant work on Spot, ideally from an SQS queue.

Spot gives up to about 90% off, and AWS can reclaim it with 2 minutes notice.

Use a load balancer, RDS Proxy, EFS, Multi-AZ, Storage Gateway or Route 53 instead of modifying the app.

The question rules out SDK changes and rewrites.

Put an SQS queue between producer and consumer and scale workers on queue depth.

The queue buffers load and survives consumer failures.

Publish to an SNS topic with an SQS queue per consumer.

Each consumer processes at its own pace and nothing is lost if one is down.

Use a FIFO queue with message group IDs.

Standard queues are at-least-once with best-effort order.

Raise the visibility timeout above the maximum processing time.

The message reappears when the timeout expires before the consumer finishes.

Set a redrive policy with maxReceiveCount to a DLQ.

Poison messages move aside for inspection instead of looping.

Model the process as a state machine, Standard for long-running, Express for high volume.

It gives built-in retries, error handling and audit history without custom code.

Enable Multi-AZ so a synchronous standby takes over behind the same endpoint.

Read replicas are asynchronous and don't fail over automatically.

Add read replicas for reporting, or ElastiCache in front for repeated reads.

Multi-AZ adds availability, not read capacity.

Use an Aurora Global Database with a secondary Region.

Storage-level replication lags by about a second and promotion takes about a minute.

Use global tables so each Region accepts writes.

It is multi-Region and multi-active with no custom replication.

Copy backups to the recovery Region and restore on demand.

It is the cheapest strategy and has the slowest recovery.

Create a VPC gateway endpoint.

It's a route table entry, free, and avoids NAT gateway charges.

Publish an endpoint service behind a Network Load Balancer.

Consumers get an interface endpoint without full network connectivity.

Attach all VPCs and VPNs to a Transit Gateway.

Peering is non-transitive and becomes a mesh.

Order Direct Connect, with a VPN as backup.

Dedicated link, steady latency, lower egress cost.

Create a Site-to-Site VPN.

Minutes to set up, IPsec encrypted, two tunnels.

Add a NACL deny rule.

Security groups can't deny.

Attach a role through an instance profile.

Temporary credentials with no keys on disk.

Apply a service control policy at the OU or root.

SCPs cap permissions across accounts and limit even root in member accounts.

Use Control Tower with controls and Account Factory.

It builds the landing zone with guardrails.

Federate the corporate IdP into IAM Identity Center with permission sets.

Central access without IAM users in each account.

Use a Cognito user pool, with an identity pool if they need AWS credentials.

User pools authenticate, identity pools issue AWS credentials.

Store them in Secrets Manager with rotation enabled.

Parameter Store has no built-in rotation.

Keep the bucket private and use OAC with a bucket policy.

Users go through the distribution only.

Mount EFS from all instances.

It is NFS, elastic and multi-AZ.

Use FSx for Windows File Server.

Native SMB and AD.

Use FSx for Lustre linked to the bucket.

Sub-millisecond latency and huge throughput.

Use io2 Block Express.

Up to 256,000 IOPS, sub-millisecond.

Use st1 throughput-optimised HDD.

Cheap for big sequential reads but can't be the boot volume.

Use instance store NVMe.

Highest IOPS but lost on stop or failure.

Enable Transfer Acceleration with multipart upload.

Enters through the nearest edge location.

Increase Lambda memory.

CPU scales with memory.

Use Intelligent-Tiering.

Automatic tiers, no retrieval fees.

Lifecycle to Deep Archive.

Cheapest storage, 180-day minimum.

Add a gateway endpoint.

It is free and bypasses the NAT gateway.

Use on-demand capacity.

Pay per request without planning.

Buy a Compute Savings Plan.

It applies across families, Regions, Fargate and Lambda.

Stop the instances on a schedule, use small sizes and Spot.

Idle hours are the waste.

Use Aurora Serverless v2.

Scales capacity with load.

Create a budget with alerts and actions.

Cost Explorer analyses, Budgets alerts.

Tag resources and activate the tags in Billing.

Reports then group by tag.