KNIGHT CAPITAL / 2012
Forty-five minutes. $460 million gone.
Start with deployment consistency and an independent way to stop harmful behavior.
Read case file 002 ↗TRAIL 01 / THREE CASES, ONE QUESTION
The change that spread.
A software release, a security rule, and a maintenance command look different on a change calendar. In an incident, the same question matters: how far can the effect travel before someone can stop it?
0 of 3 marked as read · on this browser
CLOUDFLARE / 2019
A security rule became an availability problem.
Look beyond correctness: a configuration change can exhaust the resources needed to serve traffic.
Read case file 008 ↗CROWDSTRIKE / 2024
The update that stopped the world.
Consider the recovery work left behind even after the faulty update is withdrawn.
Read case file 001 ↗THE QUESTION TO TAKE BACK
What stops your next change from reaching everyone at once?
- Which observation would stop a rollout?
- Who can halt it, and what does that control depend on?
- How do you recover devices that can no longer reach you?