system-design-primer
simulation 02a host dies mid-shift

Same traffic, but at six seconds one of your servers goes down and takes its in-flight requests with it. Survive the failure inside the error budget. N servers sized exactly for peak means N-1 is not enough.

pass when
error rate ≤ 6% · p99 ≤ 450ms · spend ≤ $500

The simulation reads best on a wider screen — the topology is drawn left to right.

Ready to run.

served
0
dropped
0
error rate
0.0%
p99
0ms
in flight
0
build
spend$100 / $500

Capacity is never free. Passing under budget is the whole exercise.