Error Budget
An error budget is the amount of unreliability permitted by a service-level objective during its evaluation window.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/error-budget/
01What it means
An error budget translates an SLO into an allowance for unsuccessful or slow events. If an objective requires 99.9% good events, up to 0.1% may be bad within the window. Teams can express the allowance as events, minutes, or another unit consistent with the underlying indicator.[1][2]
- SLO target
- Required fraction of good events
- Budget
- Permitted bad-event fraction
- Burn
- Actual bad events consumed in the window
- Burn rate
- Speed of consumption relative to allowance
Budget and consumption must use the same indicator and evaluation window; policy defines what happens as burn accelerates.
02Why it matters
The budget reframes reliability as a managed trade-off rather than a demand for perfection. Healthy remaining budget supports responsible change; rapid consumption signals that reliability work deserves priority. A shared policy can replace opinion-driven arguments between product teams seeking speed and operators seeking stability.
03When to use it
Use an error budget after establishing a valid SLI and SLO. It is especially useful for services with frequent releases or recurring tension over operational risk. Pair the number with an agreed policy describing what happens when burn is fast, the budget is nearly exhausted, or exceptions are justified.
04How it works
- Calculate the permitted bad-event fraction from the SLO target.
- Measure actual bad events over the same evaluation window.
- Track remaining budget and burn rate at useful time scales.
- Apply predefined actions, such as release review or reliability investment, when thresholds are crossed.
- Reset or roll the window as specified and review whether the policy changed behavior constructively.
05A practical example
An API with a 99.9% monthly success objective receives ten million eligible requests, allowing ten thousand failures. A faulty release causes four thousand failures in one hour. Although the monthly target has not yet been missed, the burn rate triggers the policy: rollout stops, incident response begins, and reliability fixes take precedence.
06Where it breaks
Budgets are only as honest as the SLI and objective beneath them. Long windows can conceal acute harm, while rigid release freezes can punish teams without reducing risk. Low-traffic services may need statistical care, and safety or security failures should not be treated as acceptable merely because budget remains.
Common mistakes
- Treating the entire allowance as a quota that ought to be spent.
- Watching only end-of-window compliance instead of burn rate.
- Declaring a freeze without a policy, scope, or path to recovery.