8 SAA-C03 Question Patterns People Get Wrong, and the Rule That Fixes Each One
Most wrong answers on the AWS Solutions Architect – Associate exam are not about a missing fact. They come from a handful of patterns where two options look equally good and one small detail in the stem decides. We went through the 395 SAA-C03 questions we wrote for this site and grouped the ones built around such a trap. Here are the eight patterns that come up most, with the rule that settles each one.
Common mistakes · AWS Solutions Architect – Associate (SAA-C03)

1. Multi-AZ when the problem is read load (or read replicas when it is failover)
How it shows up: "Reports run against the production database and slow down the application" next to an option that enables Multi-AZ. Or "the database must survive the loss of an Availability Zone with no data loss" next to an option that adds a read replica in another AZ.
The rule: Multi-AZ is for availability. The classic Multi-AZ standby takes over on failure and does not serve reads. Read replicas are for read scaling; replication to them is asynchronous, so promoting one can lose the last writes. Read load → replicas. AZ failure with no data loss → Multi-AZ. A good number of questions in our bank turn on this distinction.
2. Security groups and network ACLs treated as the same kind of filter
How it shows up: Traffic is allowed in but replies never arrive, or an option tries to block one IP address with a security group.
The rule: Security groups are stateful and only allow; return traffic is allowed automatically. Network ACLs are stateless, have allow and deny rules, and need explicit rules for return traffic on ephemeral ports. "Block this address" → network ACL (or AWS WAF for HTTP). "Replies are dropped" → look for a missing outbound rule on a network ACL.
3. EBS when several instances must share the files
How it shows up: An Auto Scaling group of Linux instances in several AZs needs the same uploaded files, and one option attaches an EBS volume.
The rule: An EBS volume lives in one AZ and is normally attached to one instance (Multi-Attach is limited to specific volume types in one AZ and needs a cluster-aware file system). Shared POSIX files across instances and AZs → Amazon EFS. Windows file shares → FSx for Windows File Server. Objects reached over HTTP → S3.
4. Service control policies expected to grant access
How it shows up: An option "attaches an SCP that allows S3 access" to fix an access-denied error in a member account.
The rule: An SCP only sets the maximum permissions for accounts in an organization. It never grants anything; an IAM policy (or resource policy) still has to allow the action. If the stem says "prevent any account from…", SCP is right. If it says "a user cannot…", look at IAM first. SCPs also do not apply to the management account.
Going deeper: how AWS combines policies, and why an SCP can only take away
For a request in a member account, AWS checks several policy types and the request must survive all of them: an explicit deny anywhere ends the evaluation; otherwise every applicable boundary (SCPs, permissions boundaries, session policies) must allow the action, and an identity-based or resource-based policy must also allow it. The effective permission is the intersection of the boundaries with what is granted.
An SCP is one of those boundaries. It filters, it never adds, which is why "attach an SCP that allows" never fixes an access-denied error. The same logic explains permissions boundaries on IAM roles, a favourite of delegated-administration questions.
5. KMS permissions checked only in IAM
How it shows up: A role in another account has an IAM policy that allows kms:Decrypt, but decryption still fails.
The rule: A customer managed KMS key has its own key policy, and for cross-account use both the key policy in the key's account and the IAM policy in the caller's account must allow the action. AWS managed keys cannot be shared across accounts at all, which is why "use a customer managed key" is often the first half of the answer.
6. CloudFront and Global Accelerator chosen by name, not by traffic type
How it shows up: A global multiplayer game over UDP, or an application that needs static IP addresses, with CloudFront among the options.
The rule: CloudFront is a content delivery network for HTTP(S) and caches content at the edge. Global Accelerator routes TCP and UDP traffic over the AWS network to the nearest healthy endpoint and gives two static anycast IP addresses; it does not cache. Cacheable web content → CloudFront. Non-HTTP traffic, static IPs or fast regional failover → Global Accelerator.
7. Spot Instances for work that cannot be interrupted
How it shows up: "Most cost-effective" next to a workload such as a stateful database or a job that must finish by a deadline without restarting.
The rule: Spot capacity can be taken back with a two-minute warning. It is the cheapest answer only when the work is fault-tolerant: batch jobs that checkpoint, stateless web tiers behind a load balancer, CI runners. Steady, always-on workloads → Savings Plans or Reserved Instances. A mix → Auto Scaling with a base of On-Demand or reserved capacity and Spot on top.
8. Standard SQS queues when order or exactly-once processing matters
How it shows up: Orders for the same customer must be processed in sequence, or a payment must not be charged twice, and the options offer a standard queue with more consumers.
The rule: Standard queues give at-least-once delivery and best-effort ordering. FIFO queues keep order within a message group and remove duplicates within a five-minute window, at a lower maximum throughput. "In order" or "exactly once" → FIFO with a message group ID per customer or order. "As fast as possible, order does not matter" → standard.
Beginner tip: keep a mistake log with three columns: the pattern number from this list (or "new"), the option you chose, and the word in the stem you missed. After twenty or thirty wrong answers you will see which two or three patterns are yours, and you can review those instead of everything.
How to use this list
When you get a practice question wrong, check whether it belongs to one of these patterns. If it does, the fix is the rule, not the question: write the rule down in your own words and look for it in the next set. Most people find that two or three of these patterns account for the majority of their mistakes.
- SAA-C03 exam page · All 495 SAA-C03 questions · Timed mock exam · 28-day study plan
- By domain: Design Secure Architectures · Design Resilient Architectures · Design High-Performing Architectures · Design Cost-Optimized Architectures
- Patterns 2, 4 and 5: Design Secure Architectures
- Patterns 1, 3 and 8: Design Resilient Architectures
- Pattern 7: Design Cost-Optimized Architectures