RESILIENCE TESTING

Failures

Break part of the system on purpose and see whether the service still meets its SLA, how bad it gets, and how fast it recovers. The same run is simulated healthy for comparison.

CURRENT SERVICEExample checkout API

Edit setup

Measured baseline: 2 × 4 vCPU at 40 req/s · CPU 62% · p95 420 ms
Target setup: 3 × 4 vCPU · 8 GB RAM · 120 threads
SLA thresholds: p95 ≤ 800 ms · errors ≤ 1% · CPU ≤ 80%

Play with your cluster

p95: —
CPU: —
Threads: —
Errors: —

Each change re-runs the SimPy simulation for this exact setup. Deltas compare with where you started.

Pick a failure and run it. “A node crashes” is a good first test of N+1 redundancy.