Episode 5. Backup recovery and patching. Day 5 of your 10 day sprint.

Welcome. Halfway through. Backups are beliefs until a restore proves them. Say the key lines out loud.

Say this first: An untested backup is not a backup, restore proves it.

Concepts. RMAN full plus incremental plus archivelogs, with controlfile and spfile autobackup on. Retention must match the RPO promise. Point in time recovery replays to a target time and must be rehearsed. Patches ship with a rollback owner, a restore point, and a timebox.

Scenario. Finance needs a table back as of 2 PM after a bad job. Understand: exact target time, affected objects, dependent jobs. Prepare: backup availability for that window plus a restore destination. Identify: table-level flashback versus full PITR to auxiliary; pick the smallest blast radius that meets the need. Support: recover, validate row counts with the requester, log the timeline for audit.

Say this bridge once. Say: Every patch I ship carries a rollback plan with a named owner, because uptime promises are kept during maintenance, not just during outages.

Your Day 5 lab: write a one page recovery runbook for one database with its RTO and RPO numbers, then rehearse the PITR steps aloud.
Teaser: Tomorrow, Episode 6, Exadata. Smart scan offload and the three blockers.

Second scenario. A quarterly patch window overruns. Understand: which step stalled and the rollback timebox remaining. Prepare: Datapatch log plus restore-point age. Identify: post-patch SQL regression versus patch not yet done; check completion state before rolling back. Support: finish or back out cleanly, then fix the window estimate for next quarter.

Pause and say the backup sentence again: untested is not a backup. Twice.

Close with your intro spine. Ninety seconds. From memory. Done.