← Back to recent posts
SRE

SRE Runbooks That Actually Get Used

A runbook that nobody opens at 2 a.m. is a wiki page. A runbook that gets used is a script for humans: symptom, first commands, dashboard, escalate.

Write it from the last three pages

Do not start from architecture. Start from the last PagerDuty title. If the page was “API 5xx”, the runbook title is that, not “microservices overview.”

The only sections that matter

  1. How it looks: exact metric, log line, or user report.
  2. First five minutes: three commands and one dashboard. No philosophy.
  3. Decision: restart, rollback, scale, or call the owner.
  4. When to stop: who to page next, and what “stable” means.
# First five minutes — API 5xx
kubectl -n prod get deploy api
kubectl -n prod rollout status deploy/api
kubectl -n prod logs -l app=api --tail=100
# Then the Grafana board named in the alert, not “explore.”

Keep it short

If a step needs a novel, link to a deeper doc. The on-call person is half awake. They will not read your history of the service.