Нагрузочное профилирование HTTP-сервисов

★ 6.6 · testing

http-load-profiler is a Claude Code skill that runs stepped HTTP load tests by ramping concurrency levels to collect p50/p90/p99 latency percentiles, detect performance inflection points, and recommend the optimal concurrency for a target URL. It auto-detects wrk (preferred) or ab (Apache Bench) and steps through concurrency levels — defaulting to 1 → 10 → 50 → 100 → 200 → 500 — measuring RPS, average latency, and error rates at each stage. The inflection-point algorithm flags four conditions: p99 latency accelerating beyond 2× the previous step's increase, throughput efficiency dropping more than 40% per connection, throughput saturation with a latency spike, and error rate exceeding 1% while doubling. Output is available as a human-readable table or JSON report, making it practical for API benchmarking, capacity planning, and identifying where an HTTP service saturates under load.