SLO error budgets · burn-rate alerts

Turn an SLO into alerts that page for the right reasons.

Enter the SLO and the PromQL for its SLI. You get the error budget, how fast each alert fires, and rules for Prometheus, Grafana, and Terraform. The method is the multiwindow, multi-burn-rate alert from chapter 5 of the Google SRE Workbook.

Runs in your browser. No backend, no analytics, no cookies. The whole config lives in the URL, so a link is a share.

Objective

Step 1

Service name SLO target (%) SLO window (days)

SLI type Availability: error and total counters Latency: histogram bucket at a threshold

Error events selector Total events selector

Good events: the bucket at your threshold (le) Total events: the histogram _count

Group by labels (optional, comma separated) Minimum traffic to alert (req/s, 0 turns it off) Current traffic (req/s, optional)

Each selector is a plain PromQL series selector. The tool wraps it in sum(rate(...[window])). Group-by labels go into every sum by and every and on, so each cluster or namespace burns its own budget. On Prometheus 3, whole-number bucket boundaries are stored as le="1.0", not le="1".

Alert policy

Step 2

Each row alerts when the long and the short window both burn faster than the row's rate. Burn rate = budget share × SLO window ÷ long window. The defaults are the workbook's starting point.

Budget burned (%)	Long window	Short window	Severity	Remove

Error budget

of events may fail

per day

per week

per 30 days

Downtime means a full outage: every request failing. Partial failure stretches the same budget further.

When the alerts fire

Step 3

What if errors run at a steady% from the start of the window?

Policy	Burn rate	Fires above	Full outage fires in	Budget spent by then	At the what-if rate

Budget remaining at the what-if rate

Times assume the window was clean before the errors started, and ignore scrape and rule evaluation intervals. The short window keeps an alert from firing late on old errors and lets it resolve soon after recovery.

Generated config

Prometheus rule file

Recording rules for every window, then one alert per policy row

Grafana file provisioning

Grafana-managed rules, no recording rules needed

Terraform

grafana_rule_group for the grafana/grafana provider

Grafana rules carry a for of one minute, so they fire about a minute after the Prometheus rules. Replace the runbook_url placeholders before you ship.