Backup and Restore Service · View 20 of 26 · 6 · Operations
Decisions
- The certification matrix restores real data at depth 3 for every engine version seen in the estate against every tool version in use. A pair that passes is added to a signed file in Git. Adapters read that file and refuse any pair not in it.
- Engine versions come from discovery, not from what teams say they run. The versions that cause trouble are the ones nobody planned.
- Adapter releases go to Tier 3 first, then Tier 2, then Tier 1, each after a clean rehearsal cycle. A capture tool is the one component where a subtle bug does no visible harm until it is needed.
Numbers
- A matrix run covers about 40 pairs and takes 3 to 4 hours on the rehearsal cluster, overnight. A new PostgreSQL minor version is certified the day after it is first seen.
Risks
- An owner who upgrades before certification finishes has blocked captures for up to a day, and gets paged. That is the design working: an hour of RPO-at-risk that everyone can see is better than months of copies that cannot be restored.