The balancer splits every request between two servers.

1. 59s130 requests a second, 65 to each server
2. 59sboth servers near 43% CPU, the database at 62%

app-2 crashes. Like a host dying mid-shift.

1. 60sa crash is injected into app-2: its process stops
2. 61sapp-2 answers nothing; its cables drain
3. 61sthe balancer is left with one server

Everything shifts to app-1.

1. 61sapp-1 takes all 130 requests a second
2. 61sits CPU jumps from 43% to 86%

app-1 pins at 100% CPU. The database saturates.

1. 62sthe database reaches 95% CPU
2. 63sapp-1 reaches 100% CPU: no headroom left

Each server was sized for half the load. Losing one left no room for the other half.

A fix to try: a third server or autoscaling with headroom, then replay this exact recording to compare.