Measured, not estimated. A client in a regulated industry faced an external audit of the controls around their platform. The auditors' question wasn't "do you have backups?" but "show that they work". We made it measurable: three restore tests in a single day.
The tests
- Database restore in 90 seconds: database restored to a separate cluster, data identical to the source at row level
- Point-in-time recovery to the second: in a marker test, a change made 6 seconds after the recovery target was demonstrably absent
- Full failover test in 49 minutes: all 7 VMs restored at the failover datacenter on an isolated network. The cluster formed quorum autonomously, all 9 database clusters healthy, live environment never touched
The result
| Objective | Value |
|---|---|
| RTO (target) | 4 hours |
| RTO (measured) | 49 minutes |
| RPO (databases) | ≤ 5 minutes |
Audit documentation, a runbook plus two test reports in the auditor's own format, delivered the same day.
How it works
- Three independent backup layers: encrypted VM backups to a different datacenter (tiered retention, 6+ months), application-consistent database backups with continuous WAL archiving, and all configuration as code
- Daily automated verification of every backup
- Restore testing as part of the service: database quarterly, full failover annually
A backup only exists once the restore is proven.
Read the full case study (PDF)
This recovery regime is part of our Managed Clusters and our CaffeineStacks managed servers. Same backup layers, same tested restores, on infrastructure we own. Want to be able to prove this for your platform? Get in touch.