RDS Flashcards
(12 cards)
Multi AZ
- cluster vs instance
In cluster, there is one writer and 2 reader nodes all in different AZs (max 2 readers). Application will have to implement logic to load balance usage of the 2 readers.
In instance, there is one primary and failover node. Failover node has to be in same region as primary. Failover is done by changing DNS resolution.
Both have
- Synchronous replication
Read replicas in RDS synchronous or asynchronous
Async replication. Can also live in another region.
RDS backups
Managed and manual. Stored into s3. Doesn’t auto delete, you need to delete yourself
How does RDS restore work from snapshots?
RDS creates a new instance with a new DNS name and takes time because it needs to replay every transaction from backed up transaction logs
Using read replica vs snapshots to restore
Read replica
- Can promote RR instantly
- If issue is data corruption, RR won’t help
Snapshots
- Take awhile because it needs to replay events
Aurora - difference between Aurora’s read replica to RDS’s read replica and backup node
In RDS, read replicas are not failover nodes by default (though you can manually set it to be). Failover nodes in RDS are just backups and don’t serve traffic. So you must choose between scalability and high availability in RDS.
In Aurora, read replicas are also failover nodes.
Aurora vs RDS billing
RDS
- preallocate and pay
Aurora
- high “watermark”. every month, billed the max amount of storage in GB. (will change soon)
- IO cost per request
Aurora replicated how many times by default?
- 3 per AZ and max of 15.
Aurora serverless
Same as aurora but charged by ACU which decides compute and storage. Can scale on a MIN and MAX ACU value
Aurora multi-master
Single master
- One R/W and replicas
- Failover to replicas take time
Multi master
- All instances are R/W
RDS proxy
Long running connection pooler. Use cases
- Lambda. So that connections don’t keep getting created and we can recycle connection
- Long running apps
- “Too many connections” error
Managed service
RDS custom
Gives you more control over configurations of your MS SQL and Oracle instances than RDS, but more than deploying on EC2 only