The shape of the drill
Thirty minutes, split into three parts you keep to even when the fix is close.
- Five minutes reading the symptom and writing a first hypothesis, before opening any console.
- Twenty minutes investigating and fixing, using the console and the command line as you would at work.
- Five minutes writing down what you checked, in order, and what you would check first next time.
Pick failures you have not seen
A drill on a failure you already know is a recital. Rotate across kinds of failure: access errors, delivery that silently stops, capacity, networking, and observability gaps where the problem is that nobody was told.
Use a real account, not a diagram
The console, the command line and the logs are where the time goes in a real incident. A drill that skips them trains the wrong part of the work. Use an account where nothing you break belongs to anyone else.
Review the path, not only the result
Fixed or not, the review is where the learning is. Look for the step where you changed something before you had evidence for it, and the check you made twice.