Backups: RPO and RTO
Two promises
Amazon's guide to reliable systems names two numbers a team decides in advance:
RPO, the recovery point objective: "the maximum acceptable amount of time since the last data recovery point." It is how much data the team can afford to lose.
RTO, the recovery time objective: "the maximum acceptable delay between the interruption of service and restoration of service." It is how long the game can be down.
On a timeline
The last good copy of the data is from 03:00. Disaster strikes at 14:20. The game is back at 16:20.
Everything players did between 03:00 and 14:20 is lost: 11 hours 20 minutes of data. The game was down for 2 hours.
Do the math
1last good copy 06:002disaster 09:303game back 10:15
How much data was lost, and how long was the game down?
- 3 h 30 min lost, 45 min down
- 45 min lost, 3 h 30 min down
- 4 h 15 min of each
Show the answer
3 h 30 min lost, 45 min down
Data from 06:00 to 09:30 is lost: 3 hours 30 minutes. The game was down from 09:30 to 10:15: 45 minutes.
Firestore's two tools
Point-in-time recovery (PITR) keeps "a single version per minute", "retained for 7 days". It is "disabled by default", so a team turns it on.
Scheduled backups run daily or weekly, and each can be kept for up to 14 weeks. A restore "writes the data from a backup to a new Cloud Firestore database."
Quick check
A bug quietly damaged gardens 10 days ago, and Remy only noticed today. Which tool can still help?
- A scheduled backup: PITR only reaches back 7 days
- PITR
- Neither
Show the answer
A scheduled backup: PITR only reaches back 7 days
PITR keeps 7 days. A backup kept for weeks still holds the gardens from before the bug.
A target, not an average
Amazon compares RTO with MTTR, mean time to recovery. MTTR is "a mean value taken over several" incidents. RTO is "a target, or maximum value allowed, for a single" one.
An average of 30 minutes can hide one 8-hour outage.
A backup nobody restored
Firestore warns that a restore "will take some time and must complete before the database is accessible." Nobody knows how long until they try.
Remy restores a backup into staging every month and times it, so the RTO is a measured number, not a hope.
Quick check
A backup that has never been restored still proves the team can meet its RTO.
- True
- False
Show the answer
False
Only a timed restore shows how long recovery takes, and that the backup actually works.
Pick the protection
Different protections restore to different moments.
Bolt says: Next: find the one that meets an RPO.
Fact-checked 12 October 2026
- AWS Well-Architected: Disaster recovery objectives docs.aws.amazon.com
- Cloud Firestore: Point-in-time recovery firebase.google.com
- Cloud Firestore: Back up and restore data firebase.google.com