Oracle Database@AWS Operations, High Availability, and Migration
This domain covers running the service day to day - HA topology, backups, patching, monitoring - and moving databases onto it from on-premises, OCI, or other AWS-hosted engines.
HA and backup
- RAC inside an AZ handles node and instance failure; Data Guard or Autonomous Data Guard covers AZ or region loss with a standby and controlled switchover.
- Backups are database-level settings with configurable retention. Confirm where backups land, how long a restore takes, and rehearse restore into a separate CDB rather than trusting a "completed" status.
- Patch the Exadata infrastructure on a planned schedule; use RAC rolling patching and Data Guard switchover to keep the service online.
Monitoring
Track OCI Database Management metrics, Exadata health, tablespace growth, alert-log errors, and Data Guard lag. An AWS-side alarm cannot tell you a tablespace is full, so know which layer owns each alert before an incident.
Migration
- Physical: Zero Downtime Migration with an online Data Guard path or an offline backup path - best for large on-premises or Exadata sources.
- Logical: Data Pump for schema-sized moves; GoldenGate for near-zero-downtime, cross-version, or heterogeneous cutovers.
- From Amazon RDS for Oracle: expect logical extraction (Data Pump via S3 staging, or GoldenGate) because RDS restricts SYSDBA and OS access. Reinstate grants, directories, scheduler jobs, and database links that a schema export misses.
- Sequence: assess, provision target, migrate, validate against real workload (SQL Performance Analyzer, AWR comparison), cut over with a tested rollback.
Worked example
An 8 TB E-Business Suite database on-premises must move with under four hours of downtime. Design: ZDM builds a Data Guard standby on Oracle Database@AWS over the network link and keeps it synchronized, then performs a fast switchover; AWS application servers are repointed by DNS. Reasoning: physical replication moves 8 TB without an export window, and the standby remains a rollback target.
Traps
Assuming backups replicate cross-region by default; migrating with a schema export and discovering missing grants; scheduling infrastructure patching in the same window as a DR test; cutting over while Data Guard lag is non-zero.