# Recovery Point Objective \(RPO\)

A recovery point objective sets the maximum acceptable age of recoverable data after disruption, expressing the tolerated data-loss window.

- Type: metric
- Domain: Engineering & architecture
- Level: Intermediate
- Last reviewed: 2026-10-19
- Original: [Soroush’s Product Wiki](<https://wiki.soroush.uk/topic/recovery-point-objective/>)

## What it means

Recovery point objective, or RPO, specifies how far back in time a recovery may leave the data relative to the disruption. A fifteen-minute RPO means the plan must limit potential lost updates to that window under the defined scenario. It differs from RTO, which specifies restoration time; frequent backups alone do not prove that the recoverable data meets RPO.

## Why it matters

The cost of reconstructing transactions varies greatly by workload. Explicit data-loss tolerance guides backup frequency, replication, retention, and reconciliation decisions without assuming every dataset needs the same protection.

## When to use it

Use RPO when planning recovery for stateful services and identifying authoritative records. Define it by business impact and failure scenario. For reproducible derived data, a different recovery approach may be sufficient; critical transactions may need stronger protection.

## How it works

1. Identify authoritative datasets and the consequences of losing recent updates.
2. Set a tolerated loss window for each relevant workload or dataset.
3. Choose backup, replication, and log-retention mechanisms that can support it.
4. Account for lag, failed jobs, corruption, and correlated failures.
5. Restore and reconcile data in exercises to measure the latest trustworthy recovery point.

## A practical example

A booking platform sets a five-minute RPO for reservations. Backups run every five minutes, but a failed backup and replication lag leave only a twelve-minute-old consistent restore point in an exercise. The team improves detection and protection rather than marking the schedule as compliant.

## Where it breaks

RPO is a time-based tolerance, not a guarantee of how many records will be lost. Replication can propagate corruption, and restoration may require an older consistent point. Zero-data-loss targets require carefully justified mechanisms and still need testing.

## Common mistakes

- Equating the backup schedule with demonstrated recoverability.
- Ignoring replication lag or inconsistent linked datasets.
- Confusing tolerated data loss with the time needed to restore service.

## Sources

1. [Architecture strategies for disaster recovery](<https://learn.microsoft.com/en-us/azure/well-architected/reliability/disaster-recovery>) — Microsoft · Official documentation
2. [Contingency Planning Guide for Federal Information Systems](<https://nvlpubs.nist.gov/nistpubs/Legacy/SP/nistspecialpublication800-34r1.pdf>) — National Institute of Standards and Technology · Government guidance

## Related topics

- [Disaster Recovery](<https://wiki.soroush.uk/topic/disaster-recovery/>)
- [Recovery Time Objective \(RTO\)](<https://wiki.soroush.uk/topic/recovery-time-objective/>)
- [Database Transaction](<https://wiki.soroush.uk/topic/database-transaction/>)
