Simulated environment. All schemas, metrics, and run histories are synthetic; no live Oracle instance, credentials, or customer data are used.

Oracle Secure Data Engineering Lab

Continuity

Resilience planning

Recovery objectives are only credible if they have been rehearsed. Each scenario below records a target, a runbook and the outcome of a simulated drill, including where the drill fell short. No drill was run against a real system.

Worst-case RTO
55 min

Longest modelled recovery

Worst-case RPO
15 min

Maximum tolerated data loss

Drills rehearsed
4

All within the last quarter (demo dates)

Replication channels
Lag budgets modelled on log-based CDC and physical standby patterns. GoldenGate, ODI and Data Guard are referenced as architecture concepts — nothing here is a live feed.
ChannelPatternLagBudgetOps/secConflicts 24hState
Ledger primary → reporting standbyOLTPPRD (simulated) → RPTSTB (simulated)Log-based CDC (GoldenGate-style architecture concept)4s30s (13%)12,4000
within budget
Customer master → analytics warehouseCRMPRD (simulated) → DWHPRD (simulated)Micro-batch merge (ODI-style mapping concept)1.6m5m (32%)2,1802
over budget
Telemetry rollups → regional replicaTLMPRD (simulated) → TLMDR (simulated)Asynchronous physical standby (Data Guard concept)6.9m5m (137%)31,9000
over budget
Archive vault → immutable object storeARCPRD (simulated) → Object vault (simulated)Scheduled export with checksum manifest30.3m1h (51%)6400
within budget
Backup status
Freshness measured against a fixed demo clock of 2026-08-13 06:00 UTC.

OLTPPRD full database

attention
type: full
age: 52.7h
size: 1840 GB
duration: 74m
retention: 35d
verified: yes

OLTPPRD level-1 incremental

current
type: incremental
age: 4.8h
size: 96 GB
duration: 12m
retention: 14d
verified: yes

Archived redo shipping

current
type: archivelog
age: 1.1h
size: 18 GB
duration: 3m
retention: 7d
verified: yes

DWHPRD schema logical export

attention
type: logical export
age: 79.3h
size: 310 GB
duration: 41m
retention: 30d
verified: not yet
Scenario runbooks
Expand a scenario to read the rehearsed steps.

Design principles
  • Pipelines checkpoint watermarks so a restart never double-loads or skips a window.
  • Every load is idempotent: re-running a batch converges to the same result.
  • Recovery prefers partition-level repair over full-database restore.
  • Backups are proven by restore rehearsal, never by job exit codes alone.
  • Drill notes record shortfalls openly and convert them into scheduled work.
  • Capacity headroom is tested with replayed surge volume, not extrapolated.