What Is Mean Time to Recovery (MTTR)?

Mean time to recovery (MTTR) is the average duration it takes to restore a system to normal operation after a failure, computed across many incidents.

MTTR is one of a family of reliability averages that ops teams have used for decades. It gives a high-level sense of how quickly you typically bounce back from incidents, and pairs naturally with the DORA time-to-restore metric.

How MTTR is calculated

MTTR is the total time spent in a failed or degraded state across a set of incidents, divided by the number of incidents. If you had four outages totaling two hours, your MTTR is 30 minutes. It is an average, so a single very long incident can skew it heavily.

The wider MTTx family

MTTR (recovery): average time to restore service after failure.
MTTD (detect): average time to notice that something is wrong.
MTTA (acknowledge): average time for someone to start responding.
MTBF (between failures): average uptime between incidents.

Decomposing recovery time

Recovery time is really detect + acknowledge + diagnose + fix. Breaking MTTR into its parts shows where to invest: if most of the time is in detection, improve monitoring; if it is in diagnosis, improve observability and runbooks; if it is in the fix, improve rollback tooling.

The limits of averaging

Averaging incidents has real downsides: incidents vary wildly in cause and severity, and a mean can hide a few catastrophic outages behind many trivial ones. Many teams therefore look at distributions and worst cases alongside MTTR rather than trusting the single number.

MTTR versus time to restore

MTTR and the DORA time-to-restore metric are closely related -- both measure recovery speed. DORA tends to frame it as a delivery-performance metric tied to deployments, while MTTR is a broader operational average. Used together, they give both a per-incident and an aggregate view.

Key takeaways

MTTR is the average time to restore service after a failure.
It belongs to a family of metrics (MTTD, MTTA, MTBF) that decompose reliability.
Averages hide variance, so review distributions and worst cases too.