performance-optimization（性能优化）

Performance Optimization Skill

Systematic approach to identifying and fixing performance issues.

Core Principle

Measure, don't guess. Optimization without data is guesswork.

The Cardinal Rule

❌ NEVER optimize without measuring first

Why: Premature optimization wastes time on non-issues while missing real problems.

Exception: Obvious O(n²) algorithms when O(n) alternatives exist.

Optimization Process

1. Measure Current State (Baseline)

Before touching any code, establish metrics:

Frontend Performance:

# Chrome DevTools Performance tab
# Lighthouse audit
npm run build && du -sh dist/  # Bundle size

Backend Performance:

# Add timing logs
start = Time.now
result = expensive_operation()
elapsed = Time.now - start
Logger.info("Operation took #{elapsed}ms")

Database:

# PostgreSQL
EXPLAIN ANALYZE SELECT ...;

# Check query time in logs
grep "SELECT" logs/production.log | grep "Duration:"

Metrics to capture:

• Load time / response time
• Time to interactive
• Bundle size
• Memory usage
• Query duration
• Render time

2. Profile to Find Bottlenecks

Don't guess where the problem is - profile:

Browser Profiling:

• Chrome DevTools > Performance tab
• Record interaction
• Look for long tasks (> 50ms)
• Check for layout thrashing

Server Profiling:

# Add detailed timing
defmodule Profiler do
  def measure(label, func) do
    start = System.monotonic_time(:millisecond)
    result = func.()
    elapsed = System.monotonic_time(:millisecond) - start
    Logger.info("#{label}: #{elapsed}ms")
    result
  end
end

# Use it
Profiler.measure("Database query", fn ->
  Repo.all(User)
end)

React Profiling:

# React DevTools Profiler
# Look for:
# - Unnecessary re-renders
# - Slow components (> 16ms for 60fps)
# - Large component trees

3. Identify Root Cause

Common performance issues:

Frontend:

• Large bundle size (lazy load, code split)
• Unnecessary re-renders (memoization)
• Blocking JavaScript (defer, async)
• Unoptimized images (WebP, lazy loading)
• Too many network requests (bundle, cache)
• Memory leaks (cleanup useEffect)

Backend:

• N+1 queries (preload associations)
• Missing database indexes
• Expensive computations in loops
• Synchronous external API calls
• Large JSON responses
• Inefficient algorithms

Database:

• Missing indexes
• Inefficient query structure
• Too many joins
• Fetching unnecessary columns
• No query result caching

4. Apply Targeted Optimization

One change at a time - Measure impact of each change

Frontend Optimizations

Bundle Size Reduction:

// Before: Import entire library
import _ from 'lodash'

// After: Import only what's needed
import debounce from 'lodash/debounce'

// Or: Use native alternatives
const unique = [...new Set(array)]  // Instead of _.uniq(array)

React Performance:

// Before: Re-renders on every parent render
function ChildComponent({ items }) {
  return <div>{items.map(...)}</div>
}

// After: Only re-render when items change
const ChildComponent = React.memo(function ChildComponent({ items }) {
  return <div>{items.map(...)}</div>
}, (prev, next) => prev.items === next.items)

// Before: Recreates function every render
function Parent() {
  const handleClick = () => { ... }
  return <Child onClick={handleClick} />
}

// After: Stable function reference
function Parent() {
  const handleClick = useCallback(() => { ... }, [])
  return <Child onClick={handleClick} />
}

Code Splitting:

// Before: All in main bundle
import HeavyComponent from './HeavyComponent'

// After: Lazy load when needed
const HeavyComponent = React.lazy(() => import('./HeavyComponent'))

function App() {
  return (
    <Suspense fallback={<Loading />}>
      <HeavyComponent />
    </Suspense>
  )
}

Image Optimization:

// Before: Full-size image
<img src="/hero.jpg" />

// After: Responsive, lazy-loaded
<img
  src="/hero-800w.webp"
  srcSet="/hero-400w.webp 400w, /hero-800w.webp 800w"
  loading="lazy"
  alt="Hero image"
/>

Backend Optimizations

N+1 Query Fix:

# Before: N+1 queries (1 for users + N for posts)
users = Repo.all(User)
Enum.map(users, fn user ->
  posts = Repo.all(from p in Post, where: p.user_id == ^user.id)
  {user, posts}
end)

# After: 2 queries total
users = Repo.all(User) |> Repo.preload(:posts)
Enum.map(users, fn user -> {user, user.posts} end)

Database Indexing:

-- Before: Slow query
SELECT * FROM users WHERE email = 'user@example.com';
-- Seq Scan (5000ms)

-- After: Add index
CREATE INDEX idx_users_email ON users(email);
-- Index Scan (2ms)

Caching:

# Before: Expensive calculation every request
def get_popular_posts do
  # Complex aggregation query (500ms)
  Repo.all(from p in Post, ...)
end

# After: Cache for 5 minutes
def get_popular_posts do
  Cachex.fetch(:app_cache, "popular_posts", fn ->
    result = Repo.all(from p in Post, ...)
    {:commit, result, ttl: :timer.minutes(5)}
  end)
end

Batch Processing:

# Before: Process one at a time
Enum.each(user_ids, fn id ->
  user = Repo.get(User, id)
  send_email(user)
end)

# After: Batch fetch
users = Repo.all(from u in User, where: u.id in ^user_ids)
Enum.each(users, &send_email/1)

Algorithm Optimization

Reduce Complexity:

// Before: O(n²) - nested loops
function findDuplicates(arr: number[]): number[] {
  const duplicates = []
  for (let i = 0; i < arr.length; i++) {
    for (let j = i + 1; j < arr.length; j++) {
      if (arr[i] === arr[j] && !duplicates.includes(arr[i])) {
        duplicates.push(arr[i])
      }
    }
  }
  return duplicates
}

// After: O(n) - single pass with Set
function findDuplicates(arr: number[]): number[] {
  const seen = new Set<number>()
  const duplicates = new Set<number>()

  for (const num of arr) {
    if (seen.has(num)) {
      duplicates.add(num)
    }
    seen.add(num)
  }

  return Array.from(duplicates)
}

5. Measure Impact (Proof of Work)

ALWAYS measure after optimization:

## Optimization: [What was changed]

### Before
- Load time: 3.2s
- Bundle size: 850KB
- Time to interactive: 4.1s

### Changes
- Lazy loaded HeavyComponent
- Switched to lodash-es for tree shaking
- Added React.memo to ProductList

### After
- Load time: 1.8s (-44%)
- Bundle size: 520KB (-39%)
- Time to interactive: 2.3s (-44%)

### Evidence

Before

$ npm run build dist/main.js 850.2 KB

After

$ npm run build dist/main.js 520.8 KB

Use proof-of-work skill to document evidence

6. Verify Correctness

Tests must still pass:

# Run full test suite
npm test        # Frontend
mix test        # Backend

# Manual verification
# - Feature still works
# - Edge cases handled
# - No new bugs introduced

Common Optimization Targets

Frontend Checklist

• Bundle size < 200KB (gzipped)
• First Contentful Paint < 1.5s
• Time to Interactive < 3s
• No layout shift (CLS < 0.1)
• Images optimized (WebP, lazy loading)
• Code split by route
• Unused code removed (tree shaking)
• CSS critical path optimized

Backend Checklist

• API response time < 200ms (p95)
• Database queries optimized (EXPLAIN ANALYZE)
• No N+1 queries
• Appropriate indexes exist
• Expensive operations cached
• Background jobs for slow tasks
• Connection pooling configured
• Pagination for large datasets

Database Checklist

• Indexes on frequently queried columns
• Query execution plan reviewed
• No full table scans
• Appropriate use of LIMIT
• Joins optimized (smallest table first)
• Statistics up to date (ANALYZE)

Optimization Patterns

Lazy Loading Pattern

// Route-based code splitting
const routes = [
  {
    path: '/admin',
    component: lazy(() => import('./pages/Admin'))
  },
  {
    path: '/dashboard',
    component: lazy(() => import('./pages/Dashboard'))
  }
]

Memoization Pattern

// Expensive calculation
const ExpensiveComponent = ({ data }) => {
  // Only recalculate when data changes
  const processedData = useMemo(() => {
    return data.map(item => expensiveTransform(item))
  }, [data])

  return <div>{processedData.map(...)}</div>
}

Database Query Optimization Pattern

# Instead of multiple queries
users = Repo.all(User)
posts = Repo.all(Post)
comments = Repo.all(Comment)

# Use join and preload
users =
  User
  |> join(:left, [u], p in assoc(u, :posts))
  |> join(:left, [u, p], c in assoc(p, :comments))
  |> preload([u, p, c], [posts: {p, comments: c}])
  |> Repo.all()

Anti-Patterns

❌ Optimizing the Wrong Thing

BAD: Spending hours optimizing function that runs once
GOOD: Optimize the function that runs 10,000 times per page load

Always profile first to find real bottlenecks

❌ Premature Optimization

BAD: "This might be slow, let me optimize it"
GOOD: "This IS slow (measured 500ms), let me optimize it"

❌ Micro-optimizations

BAD: Replacing `.map()` with `for` loop to save 1ms
GOOD: Reducing bundle size by 200KB to save 1000ms

Focus on high-impact optimizations

❌ Breaking Functionality for Performance

BAD: Remove feature to make it faster
GOOD: Keep feature, make implementation faster

Performance should not come at cost of correctness

❌ Optimizing Without Evidence

BAD: "I think this will be faster" [changes code]
GOOD: "Profiler shows this takes 80% of time" [measures, optimizes, measures again]

Trade-offs to Consider

Performance vs Readability:

// More readable
const result = items
  .filter(item => item.active)
  .map(item => item.name)

// Faster (one loop instead of two)
const result = []
for (const item of items) {
  if (item.active) {
    result.push(item.name)
  }
}

Question: Is the perf gain worth the readability loss? Profile first.

Performance vs Maintainability:

• Caching adds complexity
• Memoization adds memory overhead
• Code splitting adds bundle management

Always document the trade-off made

Tools & Commands

Frontend:

# Bundle analysis
npm run build -- --analyze

# Lighthouse audit
npx lighthouse https://example.com --view

# Size analysis
npx webpack-bundle-analyzer dist/stats.json

Backend:

# Database query analysis
EXPLAIN ANALYZE SELECT ...;

# Profile Elixir code
:eprof.start()
:eprof.profile(fn -> YourModule.function() end)
:eprof.stop()

Integration with Other Skills

• Use proof-of-work skill to document measurements
• Use boy-scout-rule skill while optimizing (leave better than found)
• Use simplicity-principles skill (simpler is often faster)
• Use code-reviewer skill to verify optimization quality

Remember

1. Measure first - Find real bottlenecks
2. One change at a time - Know what helped
3. Measure impact - Verify improvement
4. Preserve correctness - Tests must pass
5. Document trade-offs - Explain why

Fast code that's wrong is useless. Correct code that's fast enough is perfect.