[AWS Re:Invent 2023] Making your Amazon Aurora cluster more resilient

Resiliency

• recover from disruption
• Dynamic scale to meet demand
• Continue operating event when impaired

Resiliency - related aurora features

Backup and restore

• Point-in time restore
• Snapshots
• AWS backup
• S3 (for long-term retention data)
• On-prem 에서 Aurora 으로 마이그레이션하려면 마이그레이션 미리 snapshot/backup 실천하는 걸 추천

Exporting cluster data to S3

• Aurora cluster → S3
• object lock (not event root can change object)
• Can query using Athena
• Can use with open search

Fast clone & clone with clone (useful for multi region replication)

• original → clone1 → clone2 → …

Read replicas

• Reader with near real-time synced data becomes writer if writers shuts down

Aurora global database

• Cross-region replicas → lag depends on how far is among replicas ?
• useful for service whose users are in multiple continentals
• Failover takes about a minute; little depends on how far primary/replica is cluster is a cross-region case
• Aurora doesn’t failover with cross-region instances automatically due to lag between region;

여담

• Dehydrating replica’s cache happens after failover?
• Transit gateway or things like that sets in instances in same region behind the scene? But cross-region instances do not have that?
• multi-region is still a big deal;

Aurora serverless v2

• For traffic with rapid growth
• Auto scaling takes mili seconds and depends on how much I want to scale up/down;
• Pre-heat available? (Scaling 수행할 일시를 설정할 수 있다?)
• Not available to set zero; has minimum resource; has to stop instances if you want to set zero;

JDBC wrapper

• Driver keeps track primary/replica event
• Does not have to wait for DNS TTL
• Python has user-contributed driver similar with JDBC wrapper. → 공식 오픈소스는 없고 검색해보니 user-contributed 라이브러리도 없는 거 같다

RDS Proxy

• Multiplexing connections between applications and cluster
• Not for auto scaling;
• For holding connections; caching;
• Writer/reader endpoint available;