How long should I retain server recovery points?
Retention should cover the period in which you may need to recover a valid earlier state. The latest successful point may already contain an unnoticed deletion or corruption. Set the policy from business needs, incident-detection delay and applicable data requirements instead of choosing an arbitrary number of copies.
Before you start
- Your recovery objectives, data classification and the recovery platform’s retention behaviour.
Instructions
1. Define useful recovery windows
Identify how far back users may report missing data and how long an incident could remain undetected. Decide which recent and older points are needed for different workloads. Have the responsible data owner confirm any applicable retention or deletion obligations rather than assuming a universal legal period.
2. Understand what the tool retains
Check whether rules operate by age, count, calendar period, source or tag. Understand dependencies between full and incremental data, and how expired snapshots or object versions are removed. Retention rules should not accidentally discard the only complete recovery chain for a critical service.
3. Review the policy safely
Use the platform’s preview or report to inspect affected recovery-point identifiers before an approved deletion policy runs. Compare projected storage growth, immutability periods and retrieval needs. Periodically test an older retained point and verify that its software and decryption keys remain available.
Check the result
- The retained points cover the approved recovery windows and required dependencies.
- Deletion decisions and policy changes have a named owner and review evidence.
Troubleshooting
- The policy retains many points but storage is still growing
- Review changed-data volume, deduplication, locked versions and the tool’s separate reclamation process. Do not manually remove repository files to force space recovery.
Official references
Use the documentation for your installed version when a flag, service name or platform feature differs.