Scale This Infrastructure
The same system at four loads. At each rung: what actually gives way first, which signals say so, and what you should do about it. The correct answer at the first rung is to change nothing — a load that is not a constraint does not deserve a component.
Rung 1
100 requests/second
What breaks first
Nothing. This is the point — a single well-configured instance pair handles this comfortably, and the correct action is to change nothing.
What the signals say
- CPU under 20% on both instances.
- Database connections in single digits.
- p99 latency flat and well inside the target.
What do you do?
Commit before the verdict is shown.
Lessons behind this rung
1 / 4