So you're grinding through AWS SAA-C03 practice tests and noticing the same scenarios keep appearing. That's by design. The exam reuses architectural patterns with different wording, and once you recognize them, the answer becomes obvious. This article covers everything you need to know about aws saa-c03 exam tips. This exam tips article covers the scenario patterns I see most often across 1008+ practice questions. Each one includes what the exam is really asking and why the wrong answers look right but aren't. Let's get into it.
Pattern 1: S3 From Private Subnets: No NAT, No Internet The classic. App in private subnets needs S3 access. No NAT gateway. No internet. Answer: S3 Gateway Endpoint (free, route-table level). NAT Gateway costs money and goes through the internet. Interface Endpoints cost $0.01/hr per AZ and are for services that don't support Gateway Endpoints. When S3 is the target and the requirement is "no NAT," it's always the Gateway Endpoint.
Pattern 2: Encrypt an Existing Unencrypted RDS Instance You cannot toggle encryption on a live RDS instance. The answer is always: snapshot โ copy snapshot with KMS encryption โ restore new instance from encrypted snapshot.
Pattern 3: Mobile App Auth + Temporary AWS Credentials Amazon Cognito. User Pool for sign-up/sign-in. Identity Pool for temporary AWS credentials via STS. Never create IAM users for app consumers: they get long-term credentials and don't scale.
Pattern 4: Layer 7 Protection + TLS + Private Back-Ends ALB (TLS termination with ACM) + WAF (Layer 7 rules for SQL injection, XSS) + private subnets for EC2. Security groups alone are Layer 3/4: they can't inspect HTTP payloads. Shield is for DDoS, not application-layer rules.
Pattern 5: Temporary Reproducible Data Don't use io2 or Provisioned IOPS for easily recreated temp files. The exam tests whether you'll over-provision. Use instance store, S3, or right-sized gp3. Follow the requirement: "temporary" = cheap storage.
Pattern 6: Millisecond Key-Value at Scale DynamoDB. Partition key + sort key model + single-digit millisecond latency = DynamoDB, every time. Not Athena (analytics), not Redshift (warehouse), not ElastiCache (cache).
Pattern 7: Data Warehouse on Terabytes Redshift. Complex SQL, high concurrency, terabytes of structured data. ElastiCache is a cache. RDS is OLTP. Athena is ad hoc S3 queries. Redshift is the warehouse.
Pattern 8: SQS Duplicate Processing Standard queues = at-least-once delivery. Duplicates happen. Fix: make processing idempotent, tune visibility timeout (longer, not shorter: shorter timeouts cause more duplicates). FIFO queues are for ordering/exactly-once, not the general fix.
Pattern 9: Multi-AZ Auto-Healing Web Tier ASG across multiple AZs + ALB with health checks. Not a single large instance. Not ASG in one AZ. Multi-AZ + auto-replace on health check failure is the only resilient answer.
Pattern 10: Data Rarely Accessed After 30 Days, Keep 7 Years S3 Lifecycle Policy โ transition to Glacier or Glacier Deep Archive after 30 days. Deep Archive is ~$0.001/GB/month vs S3 Standard at $0.023/GB/month. Keeping everything in Standard works but fails the "cost-optimized" test.
Pattern 11: Global Static Content Delivery CloudFront. CDN, edge locations worldwide, low latency. Not Global Accelerator (that's for non-HTTP like gaming/IoT). Not multi-Region S3 + Route 53 latency routing (works but operationally complex).
Pattern 12: S3 Object Lock for Compliance Compliance mode WORM retention. No one: including administrators: can delete during retention. Bucket policies with deny-delete can be changed by privileged admins. Object Lock compliance mode cannot.
Pattern 13: Multiple AWS Accounts, Centralized Control AWS Organizations + SCPs. SCPs set permission boundaries for ALL IAM entities in member accounts, including root. IAM policies alone are weaker because local admins can change them. GuardDuty for centralized threat detection across accounts.
Pattern 14: Real-Time Streaming Ingestion Kinesis Data Streams. Ordered records per partition key, real-time, persistent stream. Not SQS (messaging, not streaming). Not SNS (pub/sub, not ordered stream processing).
Pattern 15: Disaster Recovery With Hours-Level RTO Backup and restore: lowest cost, fits RTO of hours. Not active-active (expensive, for minutes-level RTO). Not pilot light (more complex). Match the DR strategy to the actual RTO requirement.
Quick Numbers to Know: S3 object max: 5 TB. Bucket size: unlimited: DynamoDB item max: 400 KB: SQS message max: 256 KB: VPC CIDR max: /16 (65,536 IPs): Spot Instances: up to 90% cheaper than On-Demand, but can be interrupted: NAT Gateway: charged per hour plus per GB processed: Gateway Endpoints avoid this
| Scenario | Best AWS Service | Why | |
|
|
| | Object storage | Amazon S3 | Durable, scalable, cost-effective | | Block storage for EC2 | Amazon EBS | High-performance attached storage | | Shared file system | Amazon EFS | Multi-AZ NFS for Linux workloads | | Relational database | Amazon RDS | Managed MySQL, PostgreSQL, SQL Server | | Key-value at scale | Amazon DynamoDB | Single-digit millisecond latency | | Data warehouse | Amazon Redshift | Complex SQL analytics at scale | | In-memory cache | Amazon ElastiCache | Sub-millisecond reads | | Global CDN | Amazon CloudFront | Edge caching worldwide | | Serverless compute | AWS Lambda | No server management | | Container orchestration | Amazon ECS or EKS | Managed containers |