RDS, Aurora & DynamoDB
AWS SAA SAA-C03 · Design High-Performing Architectures
Relational databases: Amazon RDS
Amazon Relational Database Service (RDS) is a managed relational database service supporting engines including MySQL, PostgreSQL, MariaDB, Oracle, and SQL Server. “Managed” means AWS handles provisioning, patching, backups, and failover, while you keep control of the database engine, the schema, and access. RDS runs in a VPC, and a DB subnet group must span at least two Availability Zones so the database can be placed across AZs.
Two RDS features dominate the exam. Multi-AZ deployment keeps a synchronous standby copy of the database in a different Availability Zone and automatically fails over to it if the primary fails. It is a high-availability feature, not a scaling feature: the standby does not serve read traffic. Read replicas are the scaling feature: asynchronous copies of the database that serve read queries, offloading the primary. Read replicas can be created in the same Region or a different one (cross-Region read replicas also serve as a disaster recovery building block), and they can be promoted to standalone databases. The exam's classic trap is to offer a read replica when the question asks for automatic failover, or Multi-AZ when the question asks for read scaling — they solve different problems.
RDS backups come in two forms. Automated backups take a daily snapshot plus transaction logs, enabling point-in-time recovery to any second within the retention period (up to 35 days). Manual DB snapshots are taken on demand and kept until you delete them. Backups are stored in S3, and both can be used to restore a database or copy it to another Region. Storage for RDS can scale automatically with storage autoscaling, and RDS Proxy sits in front of a database to pool connections — a lifesaver for serverless applications where thousands of Lambda functions would otherwise open too many connections.
Amazon Aurora: AWS's cloud-native relational database
Amazon Aurora is AWS's own relational database engine, compatible with MySQL and PostgreSQL, built specifically for the cloud. Its storage layer automatically replicates data six ways across three Availability Zones, grows automatically up to 128 TB, and backs up continuously to S3. Aurora separates compute from storage, so you can add up to 15 read replicas with very low replication lag, and failover typically completes in seconds. Aurora Serverless v2 scales capacity automatically up and down with demand, including to zero in the original Serverless v1 design, which makes it attractive for variable or unpredictable workloads where you pay per unit of capacity used rather than for a fixed instance size. When an exam question wants the highest availability, the fastest failover, or the most read replicas for a MySQL or PostgreSQL workload, Aurora is usually the answer — provided the workload is compatible with those two engines.
NoSQL: Amazon DynamoDB
Amazon DynamoDB is a fully managed NoSQL key-value and document database. It delivers single-digit-millisecond performance at any scale, with no servers to manage. You choose between on-demand capacity (pay per request, scales automatically, best for unpredictable traffic) and provisioned capacity (you set read/write capacity units, with auto scaling, best for predictable traffic at lower cost). A DynamoDB table stores items identified by a partition key (and optionally a sort key); you query by key, and global secondary indexes support alternative query patterns. DynamoDB Streams capture a time-ordered sequence of item-level changes, which Lambda functions can process — the standard pattern for reacting to data changes. DAX (DynamoDB Accelerator) is an in-memory cache in front of DynamoDB that drops read latency to microseconds for read-heavy workloads. Global tables replicate a table across Regions with multi-active writes for applications that need low-latency access worldwide. Point-in-time recovery gives 35 days of continuous backup, and TTL (time to live) automatically deletes expired items.
The exam tests DynamoDB against RDS with a simple rule: if the data is relational, needs joins and complex queries, or requires strong transactional guarantees across many tables, use RDS or Aurora. If the data is key-value or document-shaped, needs massive scale with predictable single-digit-millisecond reads and writes, and fits a known access pattern, use DynamoDB.
Caching, warehousing, and the rest of the family
Amazon ElastiCache provides managed in-memory caching with Redis or Memcached. It sits in front of a database to absorb repeated reads, which is the exam's answer when a relational database is overloaded by read traffic and the data can tolerate being cached. Redis supports persistence, replication, and more data structures; Memcached is simpler and easier to scale horizontally. Amazon Redshift is a data warehouse for analytics (OLAP) over large datasets — the answer for complex analytical queries, not for transactional workloads. Amazon Neptune is a graph database for highly connected data like fraud detection or recommendation engines. Amazon DocumentDB is MongoDB-compatible document storage. Amazon MemoryDB is a Redis-compatible durable in-memory database for workloads that need both speed and durability. Amazon QLDB (a ledger database) is largely superseded in newer architectures. And for moving data between databases, AWS Database Migration Service (DMS) migrates databases to AWS with minimal downtime, including homogeneous and heterogeneous migrations, while AWS Schema Conversion Tool (SCT) converts schemas between different engine types.
Choosing the right database
With so many database services, the exam rewards a simple decision framework. Start with the data model: if the data is relational, with joins and complex queries across tables, choose RDS or Aurora. If the data is key-value or document-shaped with known access patterns and massive scale, choose DynamoDB. If the workload is analytics over large datasets, choose Redshift (or Athena for ad-hoc queries directly against S3). If the data is a graph of highly connected entities — fraud rings, recommendation networks, identity graphs — choose Neptune. If the workload is a MongoDB-compatible document store, choose DocumentDB. Then consider the access pattern: repeated reads of the same data call for a cache (ElastiCache or DAX) rather than a bigger database, and global low-latency access calls for global tables or read replicas in multiple Regions.
Two comparisons deserve special attention. Redshift vs Athena: Redshift is a data warehouse you provision and load, built for repeated complex analytical queries; Athena is serverless and queries data directly in S3, charging per query, which is ideal for ad-hoc analysis without managing infrastructure. ElastiCache for Redis vs Memcached: Redis supports persistence, replication, pub/sub, and rich data structures, and is the better default for most caching; Memcached is simpler, supports only simple key-value storage, and scales horizontally with ease. Finally, remember the managed-service trade-off that runs through every database question: RDS and Aurora remove the undifferentiated work of patching, backups, and failover, but you give up operating-system access and some configuration control. Running a database yourself on EC2 is almost never the exam's answer unless the question explicitly requires a feature the managed service does not offer.
Exam traps to watch for
- Confusing Multi-AZ (automatic failover, high availability) with read replicas (read scaling). They solve different problems and the exam tests the distinction directly.
- Choosing RDS for a workload that is really key-value at massive scale with millisecond latency — that is DynamoDB's shape, not a relational database's.
- Forgetting that DynamoDB on-demand vs provisioned is a cost decision: on-demand for unpredictable traffic, provisioned with auto scaling for predictable traffic.
- Adding more read replicas when the real problem is repeated identical reads — a cache (ElastiCache or DAX) is usually cheaper and faster than more database capacity.
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.
Q1.A company runs a MySQL database on Amazon RDS. The database must remain available if the primary Availability Zone fails, with automatic failover and no data loss. Which solution meets this requirement?
Explanation: Multi-AZ keeps a synchronous standby in a second AZ and fails over automatically when the primary fails, with no data loss because replication is synchronous. Option B does not help: read replicas are asynchronous, do not fail over automatically, and are a scaling feature, not an availability feature. Option A enables point-in-time recovery but requires a manual restore, with downtime. Option D is a manual backup process that also requires a restore and can lose up to an hour of data.
Exam objective: SAA-C03 Task Statement 2.2: Design highly available and/or fault-tolerant architectures
Q2.An e-commerce application uses an RDS PostgreSQL database. Read traffic has grown 10x and the primary instance is overloaded, but write traffic is unchanged. The application can tolerate slightly stale data on reads. What is the MOST cost-effective way to handle the read growth?
Explanation: Read replicas are built for this: they are asynchronous copies that serve read traffic, offloading the primary, and the application tolerates the small replication lag. Option B improves availability but the standby serves no read traffic, so it does nothing for the overload. Option C is a major re-architecture to a different data model when the workload is still relational. Option A (vertical scaling) is more expensive and has a ceiling; it also does not separate read load from write load.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q3.A gaming company needs a database for player session data: millions of key-value lookups per second, single-digit-millisecond latency, and traffic that spikes unpredictably during tournaments. Which database service is the BEST fit?
Explanation: DynamoDB is a key-value store that delivers single-digit-millisecond latency at any scale, and on-demand capacity mode scales automatically with unpredictable spikes while charging per request. Option B is a relational database that would struggle with millions of key-value lookups per second and cannot scale that way. Option D is relational and serverless, but the workload is key-value shaped, not relational. Option A is a data warehouse for analytics, not a low-latency operational store.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q4.A company runs a MySQL workload that needs up to 15 read replicas, automatic storage scaling, and failover in seconds across three Availability Zones. Which database service BEST meets these requirements?
Explanation: Aurora is built for this: its storage replicates six ways across three AZs, it supports up to 15 read replicas with low lag, storage grows automatically, and failover typically completes in seconds. Option B supports only 5 read replicas for MySQL and fails over more slowly. Option A is a NoSQL key-value store, not a MySQL-compatible relational database. Option C is an in-memory cache, not a durable relational database.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q5.A web application's RDS database is overloaded by repeated reads of the same popular product pages. The data changes infrequently and the application can tolerate cached data that is a few minutes old. Which solution MOST cost-effectively reduces the load on the database?
Explanation: The workload is repeated reads of the same data — the textbook caching case. ElastiCache absorbs those reads in memory, so the database only gets a fraction of the traffic, at much lower cost than scaling the database itself. Option A adds a standby that serves no reads. Option C is a database migration that does not address the repeated-read pattern. Option D is a backup feature with no effect on read load.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q6.A company needs to migrate its on-premises Oracle database to Amazon RDS for PostgreSQL with minimal downtime. The schema must be converted from Oracle to PostgreSQL as part of the migration. Which combination of AWS tools should the solutions architect use?
Explanation: This is a heterogeneous migration (Oracle to PostgreSQL), so two tools are needed: AWS SCT converts the schema and code between engine types, and AWS DMS replicates the data with minimal downtime. Option A is wrong because DataSync moves file data, not database contents. Option C mixes up the tools: Transfer Acceleration speeds S3 uploads, and DMS does not convert schemas. Option B is wrong on both counts: Snowball moves bulk data on a physical device, and RDS Proxy pools connections — neither converts schemas.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q7.A serverless application uses hundreds of AWS Lambda functions that each open a connection to an RDS MySQL database. The database is exhausting its available connections. Which solution BEST resolves this problem?
Explanation: RDS Proxy pools database connections and reuses them across many Lambda invocations, so hundreds of functions share a small number of real database connections — this is the purpose-built solution for serverless connection exhaustion. Option C only raises the ceiling slightly and does not fix the pattern; connection limits are tied to instance memory. Option D adds read capacity but does nothing about connection count. Option B is a full re-architecture to a different data model when the problem is connection management, not the database engine.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions
Q8.A global application needs a DynamoDB table that users in three Regions can read and write with low latency, and writes in any Region must be visible in the others. Which DynamoDB feature meets this requirement?
Explanation: Global tables replicate a DynamoDB table across Regions with multi-active writes, so each Region serves low-latency reads and writes locally and changes propagate to the other Regions. Option B captures a stream of item changes for processing but does not replicate the table. Option D is an in-memory cache that speeds up reads in one Region. Option C is a backup feature that enables restore to a point in time, not multi-Region access.
Exam objective: SAA-C03 Task Statement 3.3: Determine high-performing database solutions