antrophiclove
← Kütüphaneye dön
Kurulum
mkdir -p .claude/commands && curl -sL https://www.antrophiclove.com/api/kaynak/load-test/raw -o .claude/commands/load-test.md

Hedef dosya: .claude/commands/load-test.md

ÖnizlemeHam metin⑂ Kendi sürümünü oluştur

Run a load test against an API endpoint to measure throughput and identify breaking points.

Steps

  1. Define load test parameters:
    • Target URL and HTTP method.
    • Concurrent connections (start low, ramp up).
    • Duration of the test.
    • Request payload and headers.
  2. Select the load testing tool:
    • wrk or wrk2 for HTTP benchmarking.
    • k6 for scripted load tests.
    • ab (Apache Bench) for simple tests.
    • hey for quick Go-based load tests.
  3. Run a warm-up phase with low concurrency (10 connections, 10 seconds).
  4. Execute the main load test in stages:
    • Stage 1: Normal load (expected concurrent users).
    • Stage 2: Peak load (2x expected).
    • Stage 3: Stress test (increase until error rate > 5%).
  5. Collect metrics at each stage:
    • Requests per second (throughput).
    • Latency distribution (P50, P95, P99).
    • Error rate and error types.
    • Resource utilization (CPU, memory) if accessible.
  6. Identify the breaking point and bottleneck.
  7. Generate a report with recommendations.

Format

Load Test: <METHOD> <endpoint>

| Stage | Concurrency | RPS | P50 (ms) | P99 (ms) | Errors |
|-------|-------------|-----|----------|----------|--------|
| Normal | 10 | 500 | 20 | 85 | 0% |
| Peak | 50 | 1200 | 45 | 200 | 0.1% |
| Stress | 200 | 800 | 500 | 2000 | 5.2% |

Breaking point: ~150 concurrent connections
Bottleneck: Database connection pool exhaustion

Recommendations:
  1. Increase connection pool size from 10 to 50
  2. Add connection queuing with backpressure

Rules

  • Never run load tests against production without explicit permission.
  • Always include a warm-up phase before measuring.
  • Ramp up gradually; do not jump to maximum load immediately.
  • Record baseline metrics before the test for comparison.
  • Stop the test if error rate exceeds 10% to avoid cascading failures.