Memory hook: Object, block, file and database access patterns are different interfaces.
Must remember
- S3 stores objects identified by bucket and key. Standard supports frequent access; Intelligent-Tiering adapts eligible data to access patterns; infrequent-access and Glacier classes trade lower storage cost against retrieval, minimum-duration and access-time constraints. Lifecycle rules automate transitions and expiration.
- EBS is persistent block storage scoped to an AZ. Snapshots support backup and volume restoration. Instance store is local temporary storage whose data can be lost with the instance lifecycle or failure; it is not a durable database backup.
- EFS provides managed shared NFS file storage. FSx offers managed file systems for particular protocols and workloads, including Windows File Server, Lustre, ONTAP and OpenZFS. Storage Gateway connects on-premises applications to cloud-backed storage using supported interfaces and caching.
- AWS Backup centralises supported resource backup policies and recovery points. Replication improves access or resilience; backups recover earlier states. Test restore procedures, not only the existence of a backup.
- RDS manages relational database engines. Aurora is a cloud-designed relational database compatible with supported MySQL/PostgreSQL interfaces. A read replica scales reads; Multi-AZ supports availability. Their exact read/failover behaviour depends on the deployment type.
- DynamoDB is managed key-value/document storage. ElastiCache provides in-memory caching. Redshift targets analytical warehousing. Neptune targets graph relationships; DocumentDB targets document workloads. Choose by query and consistency needs, not merely the word database.
- Self-managed databases on EC2 give more control but more patching, backup and availability responsibility. DMS migrates/replicates data; schema-conversion tooling addresses differences between engines. Neither guarantees every application query works unchanged.
Choose under exam pressure
| Requirement | Choice and reason |
|---|---|
| Shared Linux file directory | EFS. |
| Relational transactions with managed infrastructure | RDS or Aurora. |
| Archive rarely needed records | An appropriate S3 Glacier class after checking retrieval and retention requirements. |
Traps
- S3 is not a normal block device mounted as an EBS volume.
- Lower storage price may come with retrieval fees and minimum duration.
- A cache is not automatically the authoritative durable record.
Active recall
1. Does an EC2 reboot necessarily remove an EBS volume?
No. EBS persistence and deletion settings are separate from a reboot.
2. Which database fits predictable key-based access at scale?
DynamoDB, with suitable key design.
3. What changes when a database moves from EC2 to RDS?
AWS manages more infrastructure operations; customers retain responsibility for data, access and application behaviour.
4. A file must be retrieved immediately. Can every Glacier class satisfy that?
No. Retrieval behaviour varies by class; some require a restore wait.
5. Can database migration need both DMS and schema conversion?
Yes. Data movement and engine/schema compatibility are separate concerns.