Load Test Planning
Load tests answer a decision: can we hit this SLO under this traffic model? Vanity RPS without scenarios is noise.
Workflow
- Goal: launch readiness, regression, capacity, soak, or breakpoint.
- Model traffic: users, RPS, mix of endpoints, think time, geo if relevant.
- Define SLOs as pass/fail (e.g. p95 < 300ms, error rate < 0.1%).
- Environment: prod-like data size, caching, auth; never surprise prod without approval.
- Ramp plan: warm-up → steady → (optional) stress → cool-down.
- Observability: app metrics, infra, DB; correlate with test timeline.
- Interpret: bottlenecks, saturation, not only green/red.
Output format
## Load test plan: <system>
**Goal:** …
**Traffic model:** …
**Pass/fail:** …
### Scenarios
1. …
### Ramp
…
### Tooling notes
k6/Locust/…
### Observability checklist
…
### Risks / ethics
prod impact, data safety
Rules
- Never recommend unannounced prod stress without explicit approval.
- Pass criteria must match product SLOs when they exist.
- Distinguish average vs tail latency.
- Include auth and realistic payloads; GETs-only is usually wrong.
- Document data setup and teardown.
- Do not invent tool syntax you are unsure of — describe intent.
Edge cases
- Soak: memory leaks, connection pool exhaustion over hours.
- Spike: sudden 10x; autoscaling lag.
- Write-heavy: idempotency and cleanup.