In my audit of 520 SAA-C03 questions, there are 31 that hinge on correctly distinguishing Multi-AZ from read replicas. The last time I checked our answer-distribution data, candidates were getting these 58% right — barely better than a coin toss on a two-good-option split. The reason is that the exam deliberately writes the stems to make both features sound applicable. Here is how to cut through that every time.

The 30-second mental model

Multi-AZ is a synchronous standby in a different AZ that exists only to take over when the primary fails. You cannot read from it. You cannot write to it. You cannot see it. It exists so that a failover takes ~60–120 seconds instead of a full restore.

A read replica is an asynchronous copy that you can read from. It lives to offload read traffic from the primary. It is not a failover target in the failover-automatic-in-60-seconds sense (though you can promote one manually).

Everything else about these two features is a consequence of that one distinction.

The three keywords that decide the answer

When I grade candidates' reasoning, the people who get these right are the ones who key off three words in the stem:

  1. "Availability" or "uptime" or "RTO" → you want Multi-AZ.
  2. "Read-heavy" or "offload reads" or "read performance" → you want read replicas.
  3. "Cross-region" or "disaster recovery" or "different region" → you want a cross-region read replica (which then gets promoted).

That third one is where people drop points. Multi-AZ is a single-region feature on RDS for MySQL, PostgreSQL, MariaDB, Oracle, and SQL Server. If the scenario says "recover if the entire us-east-1 region is unavailable," Multi-AZ is not the answer. The correct answer is a cross-region read replica that you promote on disaster.

What the stem will actually say

AWS rarely gives you the easy version. Here are four real patterns from the audit:

Pattern A — the availability-dressed-as-performance scenario

"A retail company runs an order-processing database on RDS for PostgreSQL. During the Cyber Monday sale, order volume spikes 4x. They need the database to remain available even if the primary instance fails. Which option provides the highest availability?"

The word "4x" is a distractor. The question is about remaining available on failure. Answer: Multi-AZ. Read replicas do not give you automatic failover, even if you scale them to 15.

Pattern B — the read-performance scenario with an availability red herring

"A social media application's database is bottlenecked on read queries during peak hours. The application currently runs on a Multi-AZ RDS for MySQL instance. How should the architecture be modified to reduce read latency?"

The "Multi-AZ" in the stem is telling you that availability is already handled. The question is about read performance. Answer: add one or more read replicas. Upgrading to a bigger instance works but is not the "best" answer by exam standards.

Pattern C — the compliance-driven cross-region scenario

"A healthcare company needs the database to be recoverable within 10 minutes if the primary AWS region becomes unavailable. The RTO is 10 minutes, RPO is 5 minutes."

Multi-AZ does not cross regions (for standard RDS engines). Answer: cross-region read replica, promoted on DR. If the RPO were 0 minutes, you would need a different pattern entirely — probably Aurora Global Database.

Pattern D — the Aurora fork

Aurora is the big exception to the mental model above. On Aurora, the Aurora replicas serve both purposes: they are read endpoints and failover targets. So if the stem is an Aurora scenario, pick Aurora replicas for both "availability" and "read scale" — one feature, two benefits.

If the exam writer wants to distinguish Aurora from RDS-classic, the stem will explicitly say "Aurora" or "Aurora PostgreSQL/MySQL." Otherwise default to RDS behavior.

The performance number you should memorize

Multi-AZ failover takes about 60–120 seconds on modern RDS engines. Read replica promotion takes 5–15 minutes depending on instance size and lag. If the stem says "RTO of 60 seconds" or "sub-minute failover," Multi-AZ is your answer. If the stem says "RTO of 15 minutes is acceptable," either is viable but Multi-AZ is cheaper — because you were going to need it for availability anyway.

The combination answer

A lot of real-world architectures use both: Multi-AZ for availability, read replicas for read scaling. The exam occasionally tests this as "all of the above" style where the right answer is "Multi-AZ + read replicas." If the stem mentions both availability and read scaling, this is the answer. People skip it because it feels like a cop-out. It is not.

What to do in the exam

When you see an RDS question:

  1. Find the goal sentence (usually the last one in the stem).
  2. Pick out which of the three keywords appears: availability, reads, or cross-region.
  3. Apply the mental model. If both appear, the answer is the combination.
  4. Check whether the scenario is Aurora or RDS-classic. Aurora collapses the two features into one.

That process takes about 40 seconds. Over 31 questions in a typical SAA-C03 form, it is the difference between passing and sending a "why did I fail" email to Pruvos support, which we get roughly 8 of a week, and this is the single most common reason.

If you want to drill these specifically, our SAA-C03 question bank tags every Multi-AZ / read replica question so you can filter down to only those 31 scenarios and practice until the pattern is automatic. Free tier lets you try the first 10 without signing up.