Blog
Notes from production.
Engineering writing about failures that recur, what they cost, and how much access software should have to production. No launch announcements until there is something to launch.
Posts
How to test an AI SRE's accuracy claims
Most AI SRE vendors quote root cause accuracy of 80 to 94 percent. Almost all of those numbers are self-reported. Here is how to check one against your own incidents before you buy.
- AI SRE
- evaluation
Most outages start with a change
Google's SRE book puts the share of outages caused by changes at roughly 70 percent. That number should shape how you detect incidents, and most alerting ignores it.
- detection
- deploys
Why postmortem action items die
Teams write good postmortems and still meet the same incident again. The problem is rarely the analysis. It is what happens to the action items after the meeting ends.
- postmortems
- incident response
Put sudoSRE in your cluster.
We are onboarding a small group of Kubernetes teams as design partners. Install takes one Helm chart, starts read-only, and is free during the partnership.