Loading…
While this loads — worth knowing
ElastiCache RedisnotElastiCache Memcached
Redis has data structures — sorted sets, counters, pub/sub. Memcached has plain keys and nothing else.
Loading…
While this loads — worth knowing
Redis has data structures — sorted sets, counters, pub/sub. Memcached has plain keys and nothing else.
Managed relational databases: six engines run for you, plus Aurora, AWS's own MySQL- and PostgreSQL-compatible engine, with standbys, replicas and backups built in.
Where RDS and Aurora takes you7 steps, all open
Amazon RDS runs a relational database for you: Db2, MariaDB, Microsoft SQL Server, MySQL, Oracle or PostgreSQL. AWS takes the backups, applies the patches, and detects and recovers from failures. You still design the schema and tune the queries.
Aurora is AWS's own engine inside RDS. It is compatible with MySQL and PostgreSQL, and keeps its data in a cluster volume copied across three Availability Zones that grows by itself. Aurora Serverless v2 adds capacity that follows the load.
A bank branch. The main vault takes every deposit, and a second vault in another building is updated with each one before the teller says done, ready to take over if the first building closes. Clerks answer balance enquiries from copies of the ledger that run a little behind.
PrimaryMulti-AZ standbyRead replicasEndpointAutomated backupsCluster volumeWrites to the primary, reports to a replica. A read replica takes heavy report queries off the primary. It is updated asynchronously, so it can run a little behind.
The standby is not a replica. In a Multi-AZ DB instance deployment the standby serves no reads at all; it is there only to take over.
No password in the code. RDS can generate the master password, keep it in Secrets Manager and rotate it, every seven days by default.
Reads scale out. Add up to 15 read replicas to a source database, each with its own endpoint that the application must choose. RDS does not add them for you. On Aurora, the reader endpoint spreads connections across the Aurora Replicas.
Writes scale up. One primary takes every write, so more write capacity means a larger instance class.
Storage grows by itself. With storage autoscaling on, RDS adds space once free space has been 10 percent or less for 5 minutes: 10 GiB, 10 percent, or the growth it predicts, whichever is most. It never shrinks.
Aurora Serverless v2 measures capacity in Aurora capacity units, adjusts it in steps as small as half a unit, and can pause at zero when nothing connects.
RDS Proxy pools connections. Many short-lived connections, such as Lambda functions', share a small pool to the database instead of each opening its own.
Multi-AZ DB instance: a primary and one standby
App serversServed by Primary
Click a node to make it fail. The timings in the log are the real ones, played 15× faster.
Aurora: a writer and one Aurora Replica, through the cluster endpoint
App serversServed by Writer
Click a node to make it fail. The timings in the log are the real ones, played 5× faster.
Where the time goes. A Multi-AZ DB instance typically fails over in 60 to 120 seconds, and Aurora typically in under 60, often under 30. How each simulation splits that between noticing and taking over is illustrative; the totals are AWS's.
The name follows the database. RDS changes the endpoint's DNS record to point at the standby, so applications reconnect to the same name and must look it up again when they do.
Aurora with no replica has nothing to promote, so it creates a new primary instance instead, typically in under 10 minutes.
What sets it off. A failed host, lost network to the primary, a storage failure, operating system patching, or a reboot you ask to fail over.
Undoing a mistake is a restore. Point-in-time restore builds a new DB instance from the backups and the transaction logs uploaded every 5 minutes; the original is left alone.
Private, behind a security group. Keep the database in private subnets, and let its security group admit only the application's.
Encryption at rest is chosen at creation. It uses a KMS key and covers the storage, logs, automated backups, read replicas and snapshots. The key cannot be changed later.
Encrypting an existing database means taking a snapshot, copying it with encryption on, and restoring from the copy.
IAM database authentication lets MariaDB, MySQL and PostgreSQL accept a token that lasts 15 minutes instead of a password, over SSL/TLS.
Secrets Manager for the master password, generated and rotated by RDS, and never shown to a person.
Instance hours, billed by the second after a 10-minute minimum. Reserved Instances lower the rate for databases that run all the time.
Storage per GB-month, plus the provisioned IOPS if you choose them.
Backup storage up to the size of the database is included; beyond that it is charged.
Replication traffic to a read replica in the same Region is not charged.
Aurora charges instance hours, or capacity-unit hours for Serverless v2, and storage per GB-month. Aurora Standard also charges for each I/O request; Aurora I/O-Optimized does not, and suits clusters whose I/O is 25 percent or more of what they spend on Aurora.
Prices change, so none are printed here. Check the RDS and Aurora pricing pages on aws.amazon.com.
| Second copy | How many, where | Serves reads | Kept in step | Takes over |
|---|---|---|---|---|
| Multi-AZ DB instance | One standby, another zone | No | Synchronously | By itself, typically 60 to 120 s |
| Multi-AZ DB cluster | Two standbys, three zones | Yes | Semisynchronously | By itself, typically under 35 s |
| Read replica | Up to 15, even in another Region | Yes | Asynchronously | Only when you promote it |
| Aurora Replica | Up to 15, on the shared volume | Yes | Usually under 100 ms behind | By itself, typically under 60 s |
| Database | Model | Grows by | Choose it when |
|---|---|---|---|
| RDS and Aurora | Relational, SQL with joins | A larger instance, and replicas for reads | Transactions and joins across tables |
| DynamoDB | Key-value and document, no joins | Itself, with no servers | Lookups by key at any scale |
| ElastiCache | In memory | More nodes or shards | Repeated reads, in front of a database |
a standby that can also serve readsA Multi-AZ DB cluster, or read replicas
A Multi-AZ DB instance's standby serves nothing until it takes over; a DB cluster's two standbys serve reads.
offload reporting queries from the primaryA read replica
It serves reads from a copy kept up to date asynchronously.
survive the loss of an Availability Zone automaticallyMulti-AZ
It fails over by itself; a read replica has to be promoted by hand.
relational disaster recovery in another Region, with a lag of about a secondAurora Global Database
It replicates to up to 10 secondary Regions, typically in under a second.
encrypt an existing unencrypted RDS databaseSnapshot, copy it encrypted, restore
Encryption at rest can only be turned on when a DB instance is created.
thousands of Lambda functions exhaust the database's connectionsRDS Proxy
It pools and shares connections, and keeps them through a failover.
keep backups for longer than 35 daysManual snapshots
Automated backups stop at 35 days; manual snapshots stay until you delete them, even after the instance is gone.
put the database back as it was 20 minutes agoPoint-in-time restore
It restores to a new DB instance, and the original is untouched.
sign in to MySQL with IAM instead of a passwordIAM database authentication
MariaDB, MySQL and PostgreSQL accept a token that lasts 15 minutes.
a relational database that is busy by day and idle at nightAurora Serverless v2
Capacity follows the load in half-unit steps, and can pause at zero.
An online shop runs its orders database on a single-AZ RDS for MySQL DB instance. If an Availability Zone fails, the database must keep running automatically without losing committed orders. Separately, heavy reporting queries are slowing the checkout. Which change meets both needs?
A web shop's app servers, in two zones, keep orders, customers and products in related tables and join them in most queries. The database must be run by AWS, keep a synchronous standby in a second zone, and fail over by itself. What do the app servers write to?
60 to 120 s.35 s.15, asynchronous, cross-Region, promoted by hand.0 to 35 days; a restore makes a new instance.256 TiB, 15 replicas.This whole page is free right now.The AWS library is still being written, so every page of it is open to everyone while that lasts. It becomes a paid bundle later; what you read today costs you nothing.
Every fact on this page was checked against AWS’s own documentation on 15 Sept 2026. If AWS has changed something since, its page is the one to trust.