Error budget calculator

An error budget is the share of requests an SLO lets fail: a 99.9% target over 10 million requests allows 10,000 failures. The burn rate is how fast you spend it compared with spending it evenly.

[BURNING FAST]

42.0% of budget used, 33.3% of the window elapsed

Allowed failures
10,000
Allowed error rate
0.1%
Budget remaining
5,800
Equivalent full downtime
43.2 min
Burn rate
1.26x
Budget runs out in
13.8 days
Projected failures at window end
12,600

Burn rate alert thresholds for a 30 day window

Burn rateLong windowShort windowBudget spentEmpty in
14.4x1 hour5 minutes2.0%2.1 days
6x6 hours30 minutes5.0%5.0 days
1x3 days6 hours10.0%30.0 days

Allowed failures per 10 million requests

SLOAllowed failuresFull downtime per 30 days
99%100,0007.2 hours
99.5%50,0003.6 hours
99.9%10,00043.2 minutes
99.95%5,00021.6 minutes
99.99%1,0004.3 minutes

How to use it

  1. Enter your SLO target as a percentage, for example 99.9, and the length of the SLO window in days. Thirty days is the usual choice.
  2. Enter the requests you expect across the whole window. Short forms like 10M or 2.5k work.
  3. Enter the failed requests counted so far and how many days of the window have passed.
  4. Read the allowed failures, the share of budget used, the burn rate and the time until the budget runs out.
  5. Use the alert table to set paging thresholds, then copy the summary into your incident or planning notes.

Questions

What is an error budget?
It is the amount of failure your SLO permits. A 99.9% success target means 0.1% of requests may fail in the window. Once those failures are used up, you have no room left for risky releases until the window rolls forward.
How is the burn rate calculated?
Burn rate is the share of the budget used divided by the share of the window that has passed. A rate of 1 spends the budget exactly by the end of the window. A rate of 2 empties it halfway through.
What burn rate should page someone?
The Google SRE workbook suggests paging at 14.4x over 1 hour (2% of a 30 day budget gone) and at 6x over 6 hours (5% gone), with a ticket at 1x over 3 days. The table above shows what share of your own window each of those spends.
Which requests count as failed?
Count what the user experiences as broken: server errors such as 500, 502, 503 and 504, timeouts and unhandled exceptions. Client errors like 404 or 401 usually stay out unless your SLO says otherwise.
Is the full downtime figure the same as the failure count?
No. It converts the budget into minutes of total outage for the window, which is how people usually talk about uptime. Partial failures spread across all traffic use up the same budget more slowly.

More free tools