← Back to Docs
Load Testing Primer
Validate your Nimbus deployment under real-world traffic before launch day.
Why Load Test
A single-threaded benchmark tells you nothing about how your licensing endpoint behaves under 10,000 concurrent activations. Load testing surfaces connection-pool exhaustion, deadlocks in your KeyAuth integration, and cold-start latency spikes before customers hit them.
Tooling
- k6 — scriptable in JavaScript, excellent for ramp-up profiles and threshold assertions.
- wrk2 — constant-throughput HTTP benchmarking with coordinated omission correction.
- vegeta — simple CLI for targeted rate attacks; pipe targets from a file.
Recipe
- Deploy a staging clone of your Vercel project with an isolated Upstash KV namespace.
- Seed 50,000 synthetic license keys via the KeyAuth API so your activation endpoint exercises the full hot path.
- Run a 5-minute ramp from 0 to 5,000 VUs against POST /api/activate.
- Assert p99 latency stays under 200ms and error rate below 0.1%.
- Repeat with a soak test at 3,000 sustained VUs for 2 hours to catch memory leaks.
Pro tip: Monitor Upstash KV read/write latency during the soak. A drifting p50 usually means your connection pool is undersized.