DeanKelly751 Load Testing
Purpose
Validate system behavior under expected and peak traffic conditions.
Workflow
Step 1: Define Load Profile
| Test type | Purpose | Load pattern |
|---|---|---|
| Smoke | Verify test works | Minimal load, short duration |
| Load | Normal conditions | Expected peak traffic, sustained |
| Stress | Find breaking point | Gradually increase until failure |
| Soak | Find memory leaks/degradation | Moderate load, long duration (hours) |
| Spike | Sudden traffic bursts | Sharp increase, then drop |
Step 2: Design Scenarios
Model real user behavior:
- Realistic request mix (not just one endpoint)
- Think time between requests (users aren't robots)
- Data variety (different payload sizes, query patterns)
- Authenticated vs anonymous traffic ratios
Step 3: Define Success Criteria
Before running:
- Maximum acceptable p99 latency
- Maximum error rate
- Minimum throughput (requests/second)
- Resource utilization ceiling (CPU <80%, memory <85%)
Step 4: Execute
- Start with smoke test (verify setup)
- Run against production-like environment
- Isolate from other traffic if possible
- Monitor system metrics during test
- Record results for comparison
Step 5: Analyze Results
- Plot latency over time (look for degradation)
- Identify saturation point (where latency spikes)
- Check error types at high load
- Compare against success criteria
- Identify the bottleneck (CPU? Memory? DB connections? Network?)
Checklist
- Load profile matches expected production traffic
- Success criteria defined before testing
- Environment is production-like (data volume, resources)
- Results baselined for future comparison
- Bottleneck identified and documented
- Capacity ceiling known (max safe traffic level)
Related Skills
performance/knowledge-optimization-principles— interpreting resultsperformance/execution-profiling— deep-dive on bottlenecks foundreviewers/review-production-readiness— load testing as readiness criterion