Management and Operations
Backup and restore semantics
Automated backups enable point-in-time recovery to any second within a retention window of 1–35 days; manual snapshots persist until you delete them and can be copied across Regions or accounts. Restoring either creates a new instance or cluster with a new endpoint - a fact exam questions exploit. Aurora Backtrack rewinds an existing cluster in place (up to 72 hours) without creating a new one, but it is not a substitute for a snapshot. DynamoDB offers PITR (35-day window) and on-demand backups, plus export to S3 that does not consume read capacity. ElastiCache backups apply to Redis/Valkey clusters, not Memcached. Redshift uses automated and manual snapshots, with cross-Region copy for DR.
Day-two operations
Maintenance windows and auto minor version upgrade control patching; EventBridge captures RDS events for alerting and automation. RDS Proxy pools and multiplexes connections, which protects against connection storms and shortens failover time for Lambda-heavy architectures. Secrets Manager rotates database credentials using a Lambda rotation function. Storage autoscaling, scaling storage or instance class, and modifying parameter groups are routine changes; most instance-class changes cause a brief outage unless done during a maintenance window.
Traps
Deleting an instance deletes its automated backups; a final snapshot is opt-in. Snapshots are incremental at the storage layer but restoring always materializes a full new resource. Backtrack cannot recover a dropped database if it happened outside the window. Copying snapshots cross-Region is required before you can restore in the DR Region.
Rule of thumb
Ask for the RTO. Minutes to hours → restore from snapshot. Seconds to a minute → Multi-AZ failover or Aurora Global Database promotion. Zero data loss plus continuous operations → synchronous Multi-AZ plus tested runbooks.