Skip to content

Observability · 6 of 6

Step 1 · The pattern

A target, and what it leaves over

The checkout API has an objective: 99% of requests succeed over 30 days. The other 1% is the error budget: failures the team has agreed it can afford. An alert rule reads the failures and decides when to call someone.

  • request
  • served
  • error budget

In short

SLO error budget burn: how it works and how it breaks

A service level objective (SLO) is a target such as: 99.9% of requests succeed over 30 days. The remaining 0.1% is the error budget, the failures you have agreed to accept. During a problem, the useful number is the burn rate: how fast the budget is being spent compared with the pace that would use it up in exactly 30 days. Burn rate 1 lasts the whole window. Burn rate 10 empties it in three days. An alert on the error rate right now reads neither. It wakes someone for a short spike that cost almost nothing, and it stays quiet for a small steady failure rate that empties the budget in days. Alert on burn rate instead, measured over a long window and a short one together. Page for a fast burn. Open a ticket for a slow one. Ignore a blip.

How this comes up in interviews

"You own a service with an objective that 99.9% of requests succeed over 30 days. How do you alert on it?"

Then they ask

  1. A small, steady failure rate sits below your error-rate threshold for days. What does your alerting do?
  2. Why measure the burn rate over a long window and a short one together?
  3. The service gets very little traffic at night. What happens to burn-rate alerts, and what do you do about it?

A strong answer covers

  • The error budget is the share of failures the objective allows over its window.
  • Burn rate is how fast the budget is being spent, compared with the pace that would use it up in exactly the window.
  • An alert on the error rate right now pages for a cheap spike and stays quiet for a slow leak.
  • Alert on burn rate, measured over a long window and a short one together.
  • Page for a fast burn, open a ticket for a slow one, ignore a blip.

Points to cover, not a script. Say them in your own words and tie them to a system you have worked on.

On one page: the track cheat sheet.

Last reviewed: Engineering write-up

Diagrams are our own simulations: the timings, rates and counts in them are illustrative, not measurements. Captions are original wording.