# Recovery time objective

## What it is

A recovery time objective, abbreviated RTO, is the target elapsed time to restore an agreed
service capability following a disruption. It answers how long the organization can plan to do
without that capability before the consequences become unacceptable. An effective objective includes
the event that starts the clock and the behavior that proves the service is useful again. Starting
from detection can omit an unnoticed outage. Stopping at a successful process start can omit a
service that customers still cannot use.

NIST's contingency planning guide describes the RTO as the maximum time a system resource may
remain unavailable before there is an unacceptable impact on other system resources, supported
business processes and the maximum tolerable downtime. The guide places recovery objectives within
the business impact analysis. A team can start with the work that needs to resume and reason back to
the components it needs. In a chain of services, the database and identity service may each need to
be ready well before the application must be usable.

The recovery point objective, or RPO, answers how far back the team can accept losing updates. A
service can be usable in twenty minutes with four hours of changes missing. It can also recover all
updates and still take half a day. Both objectives matter. A service level objective for availability
usually counts service behavior across a period, such as the proportion of good requests in a month.
An RTO addresses a defined recovery after a disruption. Several recoveries within their RTO can
still consume more of a monthly availability budget than the service owner intended.

## Why it matters

The target governs investment. A team that can tolerate a morning of downtime may be able to use a
tested restore into a fresh environment. A team that needs a usable service in minutes may need
infrastructure already in place, automated recovery checks and people available to take decisions
quickly. The elapsed time includes more than moving data. Detecting the problem, reaching a responder,
making a decision, preparing dependencies, verifying the service and switching users over can all
spend time.

A recovery target also makes the business decision precise. Visitors may tolerate a read-only
catalogue for longer than they can tolerate a failed checkout. Staff may be able to use a manual
process for a short time but fall behind as new work arrives. Defining the level of service expected
at the end of the interval makes those costs visible. Some organizations define one target for
minimum useful operation and a later milestone for clearing the backlog and restoring every function.

An objective is a plan, while the result of an exercise is an observation. A successful recovery
on a lightly loaded test system may not demonstrate that a full restore fits the same deadline. A
useful result describes the volume of data, dependencies available, who responded and the criteria
used to decide that customers could return. That lets readers judge what the exercise supports and
what it leaves unknown.

## What it looks like in practice

Imagine an order service whose owners set a two-hour recovery objective for accepting and recording
new orders. The clock starts when customers can no longer place orders. The exercise may use a test
harness to mark the interruption, so the team can measure the time spent before detection. The service
is considered restored when a set of representative orders reaches payment confirmation and appears
in the records the warehouse uses. A working catalogue alone would leave the recovery clock running.

In the exercise, detecting the outage takes twelve minutes. Declaring the incident, checking which
copy to restore and authorizing the operation take eighteen more. Restoring data takes forty-five
minutes. The team spends thirty minutes reconnecting dependencies, testing orders and opening the
route again. The observed result is 105 minutes, leaving fifteen minutes from the two-hour target.
The team separately checks the age of its newest recovered order against the target for data loss.

This exercise shows how the current procedure performed with those dependencies and that data.
Variation between attempts helps decide what to improve. If reconnecting dependencies took much
longer than expected, it may be worth preparing the configuration in advance. If nobody could
authorize the operation during one shift, the team might clarify who is allowed to make that
decision. A later exercise can test the change with a realistic amount of data and different
responders, using the same start and stop criteria.

Engineer's [backup and restore-test case](/portfolio/backups-that-have-actually-been-restored/) describes
practical verification of recovery. The company's [published service levels](/service-levels/)
discuss the terms for supported service, which can include coverage and response commitments. The
two-hour objective and timings in the order scenario are hypothetical. A reader needs an agreed
recovery objective for their own service and an exercise that observes its actual return.

## Where it goes wrong

**The clock uses a convenient timestamp.** Starting when the operator begins a restore omits the
time users already spent waiting. Agree on what counts as a disruption, collect timestamps for
detection and action, and report which interval is being measured.

**The system is called recovered when a component starts.** A healthy database or a successful
deployment can coexist with failed payments. Test the capability the target is intended to protect
through a representative user journey and its downstream records.

**Every component gets the same target.** A service may wait for a storage restore, an identity
provider and a manual decision. Give dependencies enough time within the joint timeline and allow
for work that actually has to run in sequence.

**An untested number becomes a promise.** Record the target explicitly and compare it with measured
results, including failed or slow exercises. Preserve the conditions of each test. This makes it
possible to update the plan where needed and prevents a single fast run from standing for all
outage scenarios.

## Further reading

- [NIST SP 800-34 Rev. 1](https://csrc.nist.gov/pubs/sp/800/34/r1/upd1/final) — contingency planning, business impact analysis and the relationship between recovery objectives and tolerable downtime.
- [AWS guidance on recovery objectives](https://docs.aws.amazon.com/wellarchitected/latest/reliability-pillar/rel_planning_for_recovery_objective_defined_recovery.html) — defining objectives for downtime and data loss.
- [Microsoft Azure on business continuity and recovery](https://learn.microsoft.com/en-us/azure/reliability/concept-business-continuity-high-availability-disaster-recovery) — recovery objectives and the distinctions among availability and recovery capabilities.

<https://engineer.company/glossary/recovery-time-objective/>
