Arkinu
AWS service comparison

Amazon Aurora vs Amazon RDS: differences and when to use each

Amazon Aurora
Amazon RDS

Often confused on the examQuestions tend to offer one as a distractor for the other: look for the phrase in the scenario that tips the balance.

Database

Amazon Aurora

Fully managed relational database engine compatible with MySQL and PostgreSQL that is part of Amazon RDS. It uses a distributed storage subsystem that grows automatically and automates database clustering and replication.

Exam phrases that point here

  • «MySQL- or PostgreSQL-compatible database with higher performance and availability»

    Aurora is compatible with MySQL and PostgreSQL, so existing applications work with few or no changes, and its distributed storage replicates the data across three AZs.

    When not: If the engine is Oracle, SQL Server, or Db2, Aurora does not apply (although Babelfish helps migrate from SQL Server to Aurora PostgreSQL); the answer is usually Amazon RDS. And if access is by key at massive scale with no need for relational SQL, Amazon DynamoDB is a better fit.

  • «failover in seconds with no data loss»

    Aurora Replicas share the storage volume with the writer instance, so any of them can be promoted to primary without losing data, normally in 30 seconds or less.

    When not: This covers the failure of an instance or an AZ within the Region. If you must survive the failure of an entire Region, the clue points to Aurora Global Database.

  • «up to 15 read replicas with minimal lag»

    Aurora supports up to 15 replicas per cluster and, because they read the same volume as the writer instance, the lag is usually tens of milliseconds. The reader endpoint distributes connections among them.

    When not: Replicas scale reads, not writes. If the scenario asks to scale writes beyond a single instance, other solutions come into play (such as Aurora PostgreSQL Limitless Database or DynamoDB), not adding replicas.

  • «low-latency local reads and recovery from the failure of a Region»

    Aurora Global Database replicates a cluster to other Regions with a typical lag of under a second, and, if the primary Region fails, a secondary one can be promoted to read/write within minutes.

    When not: If you only need high availability within a Region, replicas in multiple AZs are enough. If the application needs to write with low latency in several Regions at once, review the scenario: in Global Database only the primary Region writes.

  • «storage must grow automatically without provisioning it»

    The Aurora volume grows on its own with the data, up to 256 TiB, without you having to reserve storage capacity in advance.

    When not: Growing storage is not the same as scaling compute capacity. If the CPU and memory load is unpredictable, the clue points to Aurora Serverless.

Study Amazon Aurora →

Database

Amazon RDS

Managed relational database service with popular engines such as MySQL, PostgreSQL, MariaDB, Oracle Database, Microsoft SQL Server, and IBM Db2. AWS handles administration tasks such as backups, software patching, and failure detection and recovery.

Exam phrases that point here

  • «managed relational database with the least operational overhead»

    Amazon RDS handles provisioning, patching, backups, and failover for engines such as MySQL, PostgreSQL, MariaDB, Oracle, or SQL Server, without you managing the operating system.

    When not: If the scenario requires access to the operating system or an engine configuration that RDS does not allow, the option is to install the database on EC2. If it asks for high performance and storage replicated across 3 AZs with MySQL/PostgreSQL compatibility, also think of Amazon Aurora.

  • «high availability with automatic failover to another Availability Zone»

    A Multi-AZ deployment keeps a synchronous standby in another AZ and RDS automatically fails over to it if the primary or its zone fails.

    When not: Multi-AZ does not distribute reads: the standby of a Multi-AZ DB instance deployment does not serve traffic. If you also need to protect against the failure of an entire Region, you need a replica in another Region (or Aurora Global Database).

  • «offload read queries from the primary database»

    Read replicas receive an asynchronous copy of the changes and serve read-only queries, such as reports, to relieve the primary instance.

    When not: The replica can lag, so it does not work when a read must immediately see what was just written. And if the problem is that the same queries are repeated, a cache such as Amazon ElastiCache may be a better answer.

  • «copy of the database in another Region for disaster recovery»

    RDS lets you create read replicas in another Region; in a disaster, the replica is promoted to a standalone read/write instance.

    When not: Replication is asynchronous and promotion is manual, so the most recent writes can be lost. If the scenario asks for an RPO of seconds and fast recovery at global scale, it points more to Aurora Global Database; if restoring backups is enough, to copying snapshots or backups to another Region.

  • «many Lambda functions open too many database connections»

    RDS Proxy pools and reuses connections, so thousands of invocations do not exhaust the engine's connections; in addition, Lambda can authenticate with its IAM role.

    When not: If the application cannot tolerate the extra latency the proxy adds or has no connection problems, it connects directly. And if the bottleneck is reads, the solution is replicas or a cache, not a proxy.

Study Amazon RDS →