Deep dive handbook

Reliability and performance tests

Runnable example of stress test

Script and how to run

The following script, executed with tool k6 , simulates a basic stress test:

script.js

import http from 'k6/http';
import { sleep } from 'k6';

export const options = {
  stages: [
    { duration: '20s', target: 75 },
    { duration: '3m', target: 75 },
    { duration: '20s', target: 0 },
  ],
  thresholds: {
    http_req_failed: ['rate<0.01'],
    http_req_duration: ['p(95)<1000'],
  },
};

export default function () {
  const url = 'https://quickpizza.grafana.com';

  http.get(`${url}/contacts.php`);
  sleep(1);

  http.get(`${url}/news.php`);
  sleep(1);
}

Install tool k6 and download script.js , then run it locally with:

  k6 run script.js

Analysis

After being run k6 produces result .

Test summary

Stress profile:

• Up to 75 virtual users (VUs).
• 3 stages over 3m40s, followed by a 30s graceful ramp-down.
• Total duration: ~4m10s (including graceful stops).

6,656 iterations completed.

Observed behavior

• The system successfully handled up to 75 concurrent VUs.
• No request failures (http_req_failed=0,00%).
• Response times stayed well below defined thresholds throughout the test.

Metrics highlights

Requests: 13,312 total (approx. 60 requests/sec).

Response time:

• Average: 131,23ms
• Median: 130,51ms
• 95th percentile (p95): 140,19ms (well below 1000ms threshold).

Iteration duration:

• Average: 2,26s
• 95th percentile: 2,29s.

Data transferred

• Total received: 20 MB (~89 kB/sec).
• Total sent: 2,1 MB (~9,5 kB/sec).

Overall analysis

The service maintained stable performance under gradually increasing load, peaking at 75 VUs without degradation or failures. Response times remained consistent and comfortably under threshold targets, indicating solid scalability.

Suggestions and takeaways

• The system handled the stress conditions without performance issues.
• Further testing could explore higher VU limits or longer sustained periods to find actual system limits.
• Monitoring backend resource utilization (CPU, memory, DB load) during the test would provide deeper insight into system bottlenecks.