performance-engineering

# Performance Engineering

## Purpose

Performance engineering encompasses load testing, profiling, and optimization to deliver reliable, scalable systems. This skill provides frameworks for choosing the right performance testing approach (load, stress, soak, spike), profiling techniques to identify bottlenecks (CPU, memory, I/O), and optimization strategies for backend APIs, databases, and frontend applications.

Use this skill to validate system capacity before launch, detect performance regressions in CI/CD pipelines, identify and resolve bottlenecks through profiling, and optimize application responsiveness across the stack.

## When to Use This Skill

**Common Triggers:**
- "Validate API can handle expected traffic"
- "Find maximum capacity and breaking points"
- "Identify why the application is slow"
- "Detect memory leaks or resource exhaustion"
- "Optimize Core Web Vitals for SEO"
- "Set up performance testing in CI/CD"
- "Reduce cloud infrastructure costs"

**Use Cases:**
- Pre-launch capacity planning and load validation
- Post-refactor performance regression testing
- Investigating slow response times or high latency
- Detecting memory leaks in long-running services
- Optimizing database query performance
- Validating auto-scaling configuration
- Establishing performance SLOs and budgets

## Performance Testing Types

### Load Testing
Validate system behavior under expected traffic levels.

**When to use:** Pre-launch capacity planning, regression testing after refactors, validating auto-scaling.

### Stress Testing
Find system capacity limits and failure modes.

**When to use:** Capacity planning, understanding failure behavior, infrastructure sizing decisions.

### Soak Testing
Identify memory leaks, resource exhaustion, and degradation over time.

**When to use:** Detecting memory leaks, validating connection pool cleanup, testing long-running batch jobs.

### Spike Testing
Validate system response to sudden traffic spikes.

**When to use:** Validating auto-scaling, testing event-driven systems (product launches), ensuring rate limiting works.

## Quick Decision Framework

**Which test type to use?**

```
What am I trying to learn?
├─ Can my system handle expected traffic? → LOAD TEST
├─ What's the maximum capacity? → STRESS TEST
├─ Will it stay stable over time? → SOAK TEST
└─ Can it handle traffic spikes? → SPIKE TEST
```

For detailed testing patterns, load scenarios, and interpreting results, see `references/testing-types.md`.

## Load Testing Quick Starts

### k6 (JavaScript)

**Installation:**
```bash
brew install k6  # macOS
sudo apt-get install k6  # Linux
```

**Basic Load Test:**
```javascript
import http from 'k6/http';
import { check, sleep } from 'k6';

export const options = {
  stages: [
    { duration: '30s', target: 20 },
    { duration: '1m', target: 20 },
    { duration: '30s', target: 0 },
  ],
  thresholds: {
    http_req_duration: ['p(95)<500'],
    http_req_failed: ['rate<0.01'],
  },
};

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

**Run:** `k6 run script.js`

For stress, soak, and spike testing examples, see `examples/k6/`.

### Locust (Python)

**Installation:**
```bash
pip install locust
```

**Basic Load Test:**
```python
from locust import HttpUser, task, between

class WebsiteUser(HttpUser):
    wait_time = between(1, 3)
    host = "https://api.example.com"

    @task(3)
    def view_products(self):
        self.client.get("/products")

    @task(1)
    def view_product_detail(self):
        self.client.get("/products/123")
```

**Run:** `locust -f locustfile.py --headless -u 100 -r 10 --run-time 10m`

For REST API testing and data-driven testing, see `examples/locust/`.

## Profiling Quick Starts

### When to Profile

| Symptom | Profiling Type | Tool |
|---------|----------------|------|
| High CPU (>70%) | CPU Profiling | py-spy, pprof, DevTools |
| Memory growing | Memory Profiling | memory_profiler, pprof heap |
| Slow response, low CPU | I/O Profiling | Query logs, pprof block |

### Python Profiling

**py-spy (Production-Safe):**
```bash
pip install py-spy

# Profile running process
py-spy record -o profile.svg --pid <PID> --duration 30

# Top-like view
py-spy top --pid <PID>
```

**Memory Profiling:**
```python
from memory_profiler import profile

@profile
def my_function():
    a = [1] * (10 ** 6)
    return a

# Run: python -m memory_profiler script.py
```

### Go Profiling

**pprof (Built-in):**
```go
import (
    "net/http"
    _ "net/http/pprof"
)

func main() {
    go func() {
        http.ListenAndServe("localhost:6060", nil)
    }()
    startApp()
}
```

**Capture profile:**
```bash
# CPU profile (30 seconds)
go tool pprof http://localhost:6060/debug/pprof/profile?seconds=30

# Interactive analysis
(pprof) top
(pprof) web
```

### TypeScript/JavaScript Profiling

**Chrome DevTools (Browser/Node.js):**

Node.js:
```bash
node --inspect app.js
# Open chrome://inspect
# Performance tab → Record
```

**clinic.js (Node.js):**
```bash
npm install -g clinic
clinic doctor -- node app.js
```

For detailed profiling workflows and analysis, see `references/profiling-guide.md` and `examples/profiling/`.

## Optimization Strategies

### Caching

**When to cache:**
- Data queried frequently (>100 req/min)
- Data freshness tolerance (>1 minute acceptable staleness)

**Redis example:**
```python
import redis
r = redis.Redis()

def get_cached_data(key, fn, ttl=300):
    cached = r.get(key)
    if cached:
        return json.loads(cached)
    data = fn()
    r.setex(key, ttl, json.dumps(data))
    return data
```

### Database Query Optimization

**N+1 prevention:**
```python
# Bad: N+1 queries
users = User.query.all()
for user in users:
    print(user.orders)  # Separate query per user

# Good: Eager loading
users = User.query.options(joinedload(User.orders)).all()
```

**Indexing:**
```sql
CREATE INDEX idx_users_emai