Performance Testing — Load, Stress, Chaos

Performance Test Types

Type	উদ্দেশ্য
Load	Normal + peak traffic সামলে?
Stress	কখন ভাঙে? Breaking point
Spike	হঠাৎ 10x traffic
Soak/Endurance	Long duration (memory leak)
Volume	Big data set
Scalability	Resource বাড়ালে linear?

Key Metrics

• Throughput (RPS)
• Latency (p50, p95, p99)
• Error rate (%)
• Concurrent users
• Resource utilization (CPU, mem)

k6 (modern) ⭐

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

export const options = {
  stages: [
    { duration: '2m', target: 100 },   // ramp up
    { duration: '5m', target: 100 },   // stay
    { duration: '2m', target: 200 },   // spike
    { duration: '2m', target: 0 },     // ramp down
  ],
  thresholds: {
    http_req_duration: ['p(95)<500'],
    http_req_failed: ['rate<0.01'],
  }
};

export default function () {
  const res = http.get('https://api.com/users');
  check(res, { 'status 200': (r) => r.status === 200 });
  sleep(1);
}

// Run: k6 run script.js

JMeter

Apache, GUI + CLI, mature। HTTP, DB, JMS testing।

Other Tools

Gatling (Scala DSL), Locust (Python), Artillery (Node), wrk/wrk2, ab (quick), Vegeta।

Profiling ⭐

Application-এ কোথায় slow — CPU/memory bottleneck find।

• .NET — dotnet-trace, PerfView, JetBrains dotTrace
• Node — --inspect, clinic.js, 0x flamegraph
• Java — JProfiler, YourKit, async-profiler
• Python — cProfile, py-spy
• DB — EXPLAIN ANALYZE, pg_stat_statements

Flame Graph

Visual stack trace — wide function = most time। Bottleneck instant দেখা যায়।

Common Bottlenecks

• N+1 query (most common)
• Missing DB index
• Synchronous I/O (use async)
• Lock contention
• Serialization overhead
• Memory leak (long-running)
• GC pressure

Optimization Mindset ⭐

1. Measure first (profile)
2. Fix biggest bottleneck
3. Measure again
4. Repeat

Premature optimization is the root of all evil — Donald Knuth

Caching Strategies

• Cache-aside (lazy load)
• Write-through
• Write-behind (async)
• Refresh-ahead
• CDN edge cache
• Application (Redis)
• DB query cache

Chaos Engineering ⭐

Production-এ deliberate failure inject — resilience test।

Netflix Principles

1. Define steady state
2. Hypothesize কী হবে failure-এ
3. Real-world events (server kill, network lag)
4. Minimize blast radius
5. Measure, learn, fix

Tools

• Chaos Monkey (Netflix) — kill instance
• Chaos Mesh — k8s
• Gremlin — platform
• Litmus — k8s open source
• AWS Fault Injection Simulator

Game Day

Scheduled chaos — team practice incident response।

Performance Budget

Homepage budget:
  First Contentful Paint < 1.8s
  Largest Contentful Paint < 2.5s
  Total JS < 300KB
  Image < 500KB per page
  Lighthouse score > 90
// CI fail if exceeded

Web Performance ⭐

• Core Web Vitals (LCP, FID, CLS)
• Code split, lazy load
• Tree shake, minify
• Image optimize (WebP, AVIF)
• HTTP/2, HTTP/3
• CDN, brotli compression
• Preload, prefetch, preconnect
• Service Worker cache