Error budget cheat sheet
How much failure each SLO allows, over every common window.
The error budget is what your objective leaves over: 100% minus the SLO. This table converts it into time, as if the service were completely down, so you can see what each extra nine really costs.
Allowed full downtime
| SLO | 7 days | 28 days | 30 days | 90 days |
|---|---|---|---|---|
| 99% | 1 h 40 min | 6 h 43 min | 7 h 12 min | 21 h 36 min |
| 99.5% | 50 min 24 s | 3 h 21 min | 3 h 36 min | 10 h 48 min |
| 99.9% | 10 min 5 s | 40 min 19 s | 43 min 12 s | 2 h 9 min |
| 99.95% | 5 min 2 s | 20 min 10 s | 21 min 36 s | 1 h 4 min |
| 99.99% | 1 min | 4 min 2 s | 4 min 19 s | 12 min 58 s |
| 99.999% | 6 s | 24 s | 26 s | 1 min 18 s |
Rolling windows, rounded to the second. Real incidents are rarely a full outage, so the same budget lasts longer when only a share of requests fail.
Reading it with care
- Each extra nine divides the budget by ten, and the effort needed to hold it grows much faster.
- A budget of a few minutes is shorter than the time most teams need just to notice and respond to an incident.