Split View: 백엔드 성능 엔지니어링 완전 가이드 2025: 프로파일링, 부하 테스트, 병목 분석, 최적화

전체 속도 향상 = 1 / ((1 - P) + P / S)

P = 개선 가능한 부분의 비율
S = 해당 부분의 속도 향상 배수

예시: 전체의 20%를 차지하는 코드를 10배 빠르게 만들면
= 1 / ((1 - 0.2) + 0.2 / 10)
= 1 / (0.8 + 0.02)
= 1.22배 (22% 향상)

반면, 전체의 80%를 차지하는 코드를 2배 빠르게 만들면
= 1 / ((1 - 0.8) + 0.8 / 2)
= 1 / (0.2 + 0.4)
= 1.67배 (67% 향상)

# 성능 예산 정의 예시
performance_budget:
  api_endpoints:
    p50_latency_ms: 50
    p95_latency_ms: 200
    p99_latency_ms: 500
    max_latency_ms: 2000
    error_rate_percent: 0.1
    throughput_rps: 1000

  database:
    query_p95_ms: 50
    query_p99_ms: 200
    connection_pool_utilization: 70
    slow_query_threshold_ms: 100

  external_services:
    p95_latency_ms: 300
    timeout_ms: 5000
    retry_count: 3
    circuit_breaker_threshold: 50

// Node.js - 내장 프로파일러 사용
// 실행: node --prof app.js
// 분석: node --prof-process isolate-*.log > profile.txt

// 또는 v8-profiler-next 사용
const v8Profiler = require('v8-profiler-next');

function startProfiling(durationMs = 30000) {
  const title = `cpu-profile-${Date.now()}`;
  v8Profiler.startProfiling(title, true);

  setTimeout(() => {
    const profile = v8Profiler.stopProfiling(title);
    profile.export((error, result) => {
      if (!error) {
        require('fs').writeFileSync(
          `./profiles/${title}.cpuprofile`,
          result
        );
      }
      profile.delete();
    });
  }, durationMs);
}

// 미들웨어로 특정 요청 프로파일링
function profilingMiddleware(req, res, next) {
  if (req.headers['x-profile'] !== 'true') {
    return next();
  }

  const title = `req-${req.method}-${req.path}-${Date.now()}`;
  v8Profiler.startProfiling(title, true);

  const originalEnd = res.end;
  res.end = function (...args) {
    const profile = v8Profiler.stopProfiling(title);
    profile.export((error, result) => {
      if (!error) {
        require('fs').writeFileSync(
          `./profiles/${title}.cpuprofile`,
          result
        );
      }
      profile.delete();
    });
    originalEnd.apply(res, args);
  };

  next();
}

package main

import (
    "net/http"
    _ "net/http/pprof"
    "runtime"
)

func main() {
    // pprof 엔드포인트 활성화
    go func() {
        http.ListenAndServe("localhost:6060", nil)
    }()

    // CPU 프로파일 수집: go tool pprof http://localhost:6060/debug/pprof/profile?seconds=30
    // Flame Graph 생성: go tool pprof -http=:8080 profile.pb.gz

    // 또는 프로그래밍 방식으로
    // runtime.SetCPUProfileRate(100)
    // pprof.StartCPUProfile(f)
    // defer pprof.StopCPUProfile()

    runtime.SetBlockProfileRate(1)
    runtime.SetMutexProfileFraction(1)

    // 애플리케이션 로직
    startServer()
}

import cProfile
import pstats
from pyinstrument import Profiler

# cProfile 사용
def profile_with_cprofile(func):
    def wrapper(*args, **kwargs):
        profiler = cProfile.Profile()
        profiler.enable()
        result = func(*args, **kwargs)
        profiler.disable()

        stats = pstats.Stats(profiler)
        stats.sort_stats('cumulative')
        stats.print_stats(20)  # 상위 20개 함수
        return result
    return wrapper

# pyinstrument 사용 (더 읽기 쉬운 출력)
def profile_with_pyinstrument(func):
    def wrapper(*args, **kwargs):
        profiler = Profiler()
        profiler.start()
        result = func(*args, **kwargs)
        profiler.stop()
        print(profiler.output_text(unicode=True))
        return result
    return wrapper

# Django 미들웨어
class ProfilingMiddleware:
    def __init__(self, get_response):
        self.get_response = get_response

    def __call__(self, request):
        if request.META.get('HTTP_X_PROFILE') == 'true':
            profiler = Profiler()
            profiler.start()
            response = self.get_response(request)
            profiler.stop()
            response['X-Profile-Duration'] = str(profiler.last_session.duration)
            # HTML 프로파일 결과를 파일로 저장
            profiler.open_in_browser()
            return response
        return self.get_response(request)

// Node.js 힙 스냅샷
const v8 = require('v8');
const fs = require('fs');

function takeHeapSnapshot() {
  const snapshotStream = v8.writeHeapSnapshot();
  console.log(`Heap snapshot written to: ${snapshotStream}`);
  return snapshotStream;
}

// 메모리 사용량 모니터링
function monitorMemory(intervalMs = 5000) {
  setInterval(() => {
    const usage = process.memoryUsage();
    console.log({
      rss_mb: Math.round(usage.rss / 1024 / 1024),
      heapTotal_mb: Math.round(usage.heapTotal / 1024 / 1024),
      heapUsed_mb: Math.round(usage.heapUsed / 1024 / 1024),
      external_mb: Math.round(usage.external / 1024 / 1024),
      arrayBuffers_mb: Math.round(usage.arrayBuffers / 1024 / 1024)
    });
  }, intervalMs);
}

// 메모리 누수 감지 패턴
class MemoryLeakDetector {
  constructor(options = {}) {
    this.samples = [];
    this.maxSamples = options.maxSamples || 60;
    this.threshold = options.thresholdMB || 50;
  }

  sample() {
    const usage = process.memoryUsage();
    this.samples.push({
      timestamp: Date.now(),
      heapUsed: usage.heapUsed
    });

    if (this.samples.length > this.maxSamples) {
      this.samples.shift();
    }

    return this.detectLeak();
  }

  detectLeak() {
    if (this.samples.length < 10) return null;

    const first = this.samples[0].heapUsed;
    const last = this.samples[this.samples.length - 1].heapUsed;
    const diffMB = (last - first) / 1024 / 1024;

    // 지속적인 메모리 증가 패턴 감지
    let increasing = 0;
    for (let i = 1; i < this.samples.length; i++) {
      if (this.samples[i].heapUsed > this.samples[i - 1].heapUsed) {
        increasing++;
      }
    }

    const increaseRatio = increasing / (this.samples.length - 1);

    if (diffMB > this.threshold && increaseRatio > 0.7) {
      return {
        suspected: true,
        growthMB: diffMB.toFixed(2),
        increaseRatio: increaseRatio.toFixed(2),
        duration: this.samples[this.samples.length - 1].timestamp - this.samples[0].timestamp
      };
    }

    return null;
  }
}

# Python - I/O 프로파일링
import time
import functools
import logging
from contextlib import contextmanager

logger = logging.getLogger('io_profiler')

class IOProfiler:
    """I/O 작업 시간 측정 데코레이터 및 컨텍스트 매니저"""

    _stats = {}

    @classmethod
    def track(cls, operation_name):
        def decorator(func):
            @functools.wraps(func)
            async def async_wrapper(*args, **kwargs):
                start = time.perf_counter()
                try:
                    result = await func(*args, **kwargs)
                    duration = time.perf_counter() - start
                    cls._record(operation_name, duration, success=True)
                    return result
                except Exception as e:
                    duration = time.perf_counter() - start
                    cls._record(operation_name, duration, success=False)
                    raise

            @functools.wraps(func)
            def sync_wrapper(*args, **kwargs):
                start = time.perf_counter()
                try:
                    result = func(*args, **kwargs)
                    duration = time.perf_counter() - start
                    cls._record(operation_name, duration, success=True)
                    return result
                except Exception as e:
                    duration = time.perf_counter() - start
                    cls._record(operation_name, duration, success=False)
                    raise

            import asyncio
            if asyncio.iscoroutinefunction(func):
                return async_wrapper
            return sync_wrapper
        return decorator

    @classmethod
    def _record(cls, name, duration, success):
        if name not in cls._stats:
            cls._stats[name] = {
                'count': 0, 'total_time': 0,
                'min_time': float('inf'), 'max_time': 0,
                'errors': 0
            }
        stats = cls._stats[name]
        stats['count'] += 1
        stats['total_time'] += duration
        stats['min_time'] = min(stats['min_time'], duration)
        stats['max_time'] = max(stats['max_time'], duration)
        if not success:
            stats['errors'] += 1

    @classmethod
    def report(cls):
        for name, stats in sorted(cls._stats.items()):
            avg = stats['total_time'] / stats['count'] if stats['count'] else 0
            logger.info(
                f"{name}: count={stats['count']}, "
                f"avg={avg*1000:.1f}ms, "
                f"min={stats['min_time']*1000:.1f}ms, "
                f"max={stats['max_time']*1000:.1f}ms, "
                f"errors={stats['errors']}"
            )


# 사용 예시
class UserRepository:
    @IOProfiler.track('db.users.find_by_id')
    async def find_by_id(self, user_id):
        return await self.db.users.find_one({"_id": user_id})

    @IOProfiler.track('db.users.search')
    async def search(self, query, limit=20):
        return await self.db.users.find(query).limit(limit).to_list(limit)

class ExternalAPIClient:
    @IOProfiler.track('api.payment.charge')
    async def charge(self, amount, token):
        async with self.session.post('/charge', json={"amount": amount, "token": token}) as resp:
            return await resp.json()

import http from 'k6/http';
import { check, sleep, group } from 'k6';
import { Rate, Trend, Counter } from 'k6/metrics';

// 커스텀 메트릭
const errorRate = new Rate('errors');
const apiDuration = new Trend('api_duration', true);
const requestCount = new Counter('requests');

// 테스트 옵션
export const options = {
  scenarios: {
    // 시나리오 1: 일반 부하 테스트
    normal_load: {
      executor: 'ramping-vus',
      startVUs: 0,
      stages: [
        { duration: '2m', target: 50 },   // 2분간 50 VU까지 증가
        { duration: '5m', target: 50 },   // 5분간 50 VU 유지
        { duration: '2m', target: 100 },  // 2분간 100 VU까지 증가
        { duration: '5m', target: 100 },  // 5분간 100 VU 유지
        { duration: '2m', target: 0 },    // 2분간 0으로 감소
      ],
    },
    // 시나리오 2: 스파이크 테스트
    spike_test: {
      executor: 'ramping-vus',
      startVUs: 0,
      startTime: '16m',
      stages: [
        { duration: '10s', target: 500 },  // 급격한 스파이크
        { duration: '1m', target: 500 },   // 유지
        { duration: '10s', target: 0 },    // 급격한 감소
      ],
    },
  },
  thresholds: {
    http_req_duration: ['p(95)<200', 'p(99)<500'],
    errors: ['rate<0.01'],
    http_req_failed: ['rate<0.01'],
  },
};

const BASE_URL = __ENV.BASE_URL || 'http://localhost:3000';

export default function () {
  const authToken = login();

  group('API Operations', () => {
    group('List Products', () => {
      const res = http.get(`${BASE_URL}/api/products?page=1&limit=20`, {
        headers: { Authorization: `Bearer ${authToken}` },
        tags: { name: 'GET /api/products' },
      });

      check(res, {
        'status is 200': (r) => r.status === 200,
        'response time OK': (r) => r.timings.duration < 200,
        'has products': (r) => JSON.parse(r.body).data.length > 0,
      });

      errorRate.add(res.status !== 200);
      apiDuration.add(res.timings.duration);
      requestCount.add(1);
    });

    group('Get Product Detail', () => {
      const productId = Math.floor(Math.random() * 1000) + 1;
      const res = http.get(`${BASE_URL}/api/products/${productId}`, {
        headers: { Authorization: `Bearer ${authToken}` },
        tags: { name: 'GET /api/products/:id' },
      });

      check(res, {
        'status is 200': (r) => r.status === 200,
        'has product data': (r) => {
          const body = JSON.parse(r.body);
          return body.data && body.data.id;
        },
      });

      errorRate.add(res.status !== 200);
      apiDuration.add(res.timings.duration);
    });

    group('Create Order', () => {
      const payload = JSON.stringify({
        productId: Math.floor(Math.random() * 1000) + 1,
        quantity: Math.floor(Math.random() * 5) + 1,
        shippingAddress: '123 Test Street',
      });

      const res = http.post(`${BASE_URL}/api/orders`, payload, {
        headers: {
          Authorization: `Bearer ${authToken}`,
          'Content-Type': 'application/json',
        },
        tags: { name: 'POST /api/orders' },
      });

      check(res, {
        'order created': (r) => r.status === 201,
        'has order id': (r) => JSON.parse(r.body).data.orderId,
      });

      errorRate.add(res.status !== 201);
      apiDuration.add(res.timings.duration);
    });
  });

  sleep(Math.random() * 3 + 1); // 1-4초 사이 대기
}

function login() {
  const res = http.post(`${BASE_URL}/api/auth/login`, JSON.stringify({
    email: `user${__VU}@test.com`,
    password: 'testpassword',
  }), {
    headers: { 'Content-Type': 'application/json' },
    tags: { name: 'POST /api/auth/login' },
  });

  return res.status === 200 ? JSON.parse(res.body).token : '';
}

# artillery-config.yml
config:
  target: "http://localhost:3000"
  phases:
    - duration: 120
      arrivalRate: 10
      name: "Warm up"
    - duration: 300
      arrivalRate: 50
      name: "Normal load"
    - duration: 120
      arrivalRate: 100
      name: "Peak load"
  defaults:
    headers:
      Content-Type: "application/json"
  plugins:
    expect: {}
    metrics-by-endpoint: {}
  ensure:
    thresholds:
      - http.response_time.p95: 200
      - http.response_time.p99: 500

scenarios:
  - name: "User browsing flow"
    weight: 70
    flow:
      - post:
          url: "/api/auth/login"
          json:
            email: "user@test.com"
            password: "password123"
          capture:
            - json: "$.token"
              as: "authToken"
          expect:
            - statusCode: 200
      - get:
          url: "/api/products?page=1&limit=20"
          headers:
            Authorization: "Bearer {{ authToken }}"
          expect:
            - statusCode: 200
            - hasProperty: "data"
      - think: 2
      - get:
          url: "/api/products/{{ $randomNumber(1, 1000) }}"
          headers:
            Authorization: "Bearer {{ authToken }}"
          expect:
            - statusCode: 200

  - name: "Order creation flow"
    weight: 30
    flow:
      - post:
          url: "/api/auth/login"
          json:
            email: "buyer@test.com"
            password: "password123"
          capture:
            - json: "$.token"
              as: "authToken"
      - post:
          url: "/api/orders"
          headers:
            Authorization: "Bearer {{ authToken }}"
          json:
            productId: "{{ $randomNumber(1, 100) }}"
            quantity: "{{ $randomNumber(1, 5) }}"
          expect:
            - statusCode: 201

# RED Method 모니터링 구현
from prometheus_client import Counter, Histogram, Gauge
import time

# Rate: 초당 요청 수
request_count = Counter(
    'http_requests_total',
    'Total HTTP requests',
    ['method', 'endpoint', 'status']
)

# Errors: 에러 비율
error_count = Counter(
    'http_errors_total',
    'Total HTTP errors',
    ['method', 'endpoint', 'error_type']
)

# Duration: 응답 시간 분포
request_duration = Histogram(
    'http_request_duration_seconds',
    'HTTP request duration',
    ['method', 'endpoint'],
    buckets=[0.005, 0.01, 0.025, 0.05, 0.1, 0.25, 0.5, 1.0, 2.5, 5.0, 10.0]
)

# 미들웨어 구현
class REDMetricsMiddleware:
    def __init__(self, app):
        self.app = app

    async def __call__(self, scope, receive, send):
        if scope['type'] != 'http':
            return await self.app(scope, receive, send)

        method = scope.get('method', 'UNKNOWN')
        path = scope.get('path', '/')
        status_code = 500

        start = time.perf_counter()
        try:
            # 응답 상태 코드 캡처
            async def send_wrapper(message):
                nonlocal status_code
                if message['type'] == 'http.response.start':
                    status_code = message['status']
                await send(message)

            await self.app(scope, receive, send_wrapper)
        except Exception as e:
            error_count.labels(method=method, endpoint=path, error_type=type(e).__name__).inc()
            raise
        finally:
            duration = time.perf_counter() - start
            request_count.labels(method=method, endpoint=path, status=str(status_code)).inc()
            request_duration.labels(method=method, endpoint=path).observe(duration)

            if status_code >= 400:
                error_count.labels(method=method, endpoint=path, error_type=f'http_{status_code}').inc()

              평균(Mean)   p50    p95    p99    p99.9   Max
사용자 영향도   낮음       중간    높음    높음    매우높음  극단적

p50 (중앙값): 50%의 요청이 이 시간 이내에 완료
p95: 95%의 요청이 이 시간 이내에 완료 (20개 중 1개가 이보다 느림)
p99: 99%의 요청이 이 시간 이내에 완료 (100개 중 1개가 이보다 느림)

왜 평균은 위험한가?
- 평균 50ms여도 p99가 5000ms일 수 있음
- 매 100번째 요청마다 사용자가 5초를 대기
- 헤비 유저일수록 높은 백분위에 노출될 확률 증가

# BAD: N+1 쿼리 - 주문 100개면 101번 쿼리 실행
orders = Order.objects.all()[:100]
for order in orders:
    # 각 주문마다 별도 쿼리로 사용자 정보 조회
    print(f"Order {order.id} by {order.user.name}")

# GOOD: Eager loading - 2번의 쿼리로 해결
orders = Order.objects.select_related('user').all()[:100]
for order in orders:
    print(f"Order {order.id} by {order.user.name}")

# GOOD: Prefetch (M:N 관계)
orders = Order.objects.prefetch_related('items__product').all()[:100]
for order in orders:
    for item in order.items.all():
        print(f"  - {item.product.name}")

// Node.js + Prisma - N+1 해결
// BAD: N+1
const orders = await prisma.order.findMany({ take: 100 });
for (const order of orders) {
  const user = await prisma.user.findUnique({
    where: { id: order.userId }
  });
}

// GOOD: Include (Join)
const orders = await prisma.order.findMany({
  take: 100,
  include: {
    user: true,
    items: {
      include: { product: true }
    }
  }
});

// GOOD: DataLoader 패턴
const DataLoader = require('dataloader');

const userLoader = new DataLoader(async (userIds) => {
  const users = await prisma.user.findMany({
    where: { id: { in: [...userIds] } }
  });
  const userMap = new Map(users.map(u => [u.id, u]));
  return userIds.map(id => userMap.get(id));
});

// 여러 번 호출해도 자동으로 배치 처리
const user1 = await userLoader.load(1);
const user2 = await userLoader.load(2);

-- 느린 쿼리 탐지 (PostgreSQL)
SELECT
  query,
  calls,
  mean_exec_time,
  total_exec_time,
  rows
FROM pg_stat_statements
ORDER BY mean_exec_time DESC
LIMIT 20;

-- 실행 계획 분석
EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT)
SELECT o.*, u.name
FROM orders o
JOIN users u ON o.user_id = u.id
WHERE o.status = 'pending'
  AND o.created_at > NOW() - INTERVAL '7 days'
ORDER BY o.created_at DESC
LIMIT 50;

-- 복합 인덱스 생성 (쿼리 패턴에 맞게)
CREATE INDEX CONCURRENTLY idx_orders_status_created
ON orders (status, created_at DESC)
WHERE status IN ('pending', 'processing');

-- 인덱스 사용률 확인
SELECT
  schemaname,
  tablename,
  indexname,
  idx_scan,
  idx_tup_read,
  idx_tup_fetch
FROM pg_stat_user_indexes
ORDER BY idx_scan ASC;

// HikariCP 최적 설정 (Java/Spring Boot)
// application.yml
/*
spring:
  datasource:
    hikari:
      maximum-pool-size: 20
      minimum-idle: 5
      idle-timeout: 300000
      max-lifetime: 600000
      connection-timeout: 30000
      leak-detection-threshold: 60000
      pool-name: "MainPool"
*/

// 커넥션 풀 모니터링
import com.zaxxer.hikari.HikariDataSource;
import com.zaxxer.hikari.HikariPoolMXBean;

public class ConnectionPoolMonitor {
    private final HikariDataSource dataSource;

    public PoolStats getStats() {
        HikariPoolMXBean poolBean = dataSource.getHikariPoolMXBean();
        return new PoolStats(
            poolBean.getTotalConnections(),
            poolBean.getActiveConnections(),
            poolBean.getIdleConnections(),
            poolBean.getThreadsAwaitingConnection()
        );
    }

    public void logWarningIfNeeded() {
        PoolStats stats = getStats();
        double utilization = (double) stats.active / stats.total;

        if (utilization > 0.8) {
            log.warn("Connection pool utilization HIGH: {}% ({}/{})",
                Math.round(utilization * 100),
                stats.active, stats.total);
        }

        if (stats.waiting > 0) {
            log.error("Threads waiting for connection: {}", stats.waiting);
        }
    }
}

# PgBouncer 설정 (PostgreSQL 커넥션 풀러)
# pgbouncer.ini
"""
[databases]
mydb = host=127.0.0.1 port=5432 dbname=mydb

[pgbouncer]
listen_port = 6432
listen_addr = 0.0.0.0
auth_type = md5
auth_file = /etc/pgbouncer/userlist.txt

pool_mode = transaction
default_pool_size = 25
min_pool_size = 5
reserve_pool_size = 5
reserve_pool_timeout = 3
max_client_conn = 1000
max_db_connections = 50

server_idle_timeout = 600
server_lifetime = 3600
client_idle_timeout = 0

log_connections = 1
log_disconnections = 1
log_pooler_errors = 1
stats_period = 60
"""

// Go - 잠금 경합 프로파일링
package main

import (
    "runtime"
    "sync"
    "time"
)

// BAD: 글로벌 뮤텍스로 전체 맵 잠금
type BadCache struct {
    mu    sync.Mutex
    items map[string]interface{}
}

func (c *BadCache) Get(key string) interface{} {
    c.mu.Lock()
    defer c.mu.Unlock()
    return c.items[key]
}

// GOOD: 샤딩으로 잠금 경합 분산
type ShardedCache struct {
    shards    [256]shard
    shardMask uint8
}

type shard struct {
    mu    sync.RWMutex
    items map[string]interface{}
}

func NewShardedCache() *ShardedCache {
    c := &ShardedCache{shardMask: 255}
    for i := range c.shards {
        c.shards[i].items = make(map[string]interface{})
    }
    return c
}

func (c *ShardedCache) getShard(key string) *shard {
    hash := fnv32(key)
    return &c.shards[hash&uint32(c.shardMask)]
}

func (c *ShardedCache) Get(key string) (interface{}, bool) {
    s := c.getShard(key)
    s.mu.RLock()
    defer s.mu.RUnlock()
    val, ok := s.items[key]
    return val, ok
}

func (c *ShardedCache) Set(key string, value interface{}) {
    s := c.getShard(key)
    s.mu.Lock()
    defer s.mu.Unlock()
    s.items[key] = value
}

func fnv32(key string) uint32 {
    hash := uint32(2166136261)
    for i := 0; i < len(key); i++ {
        hash *= 16777619
        hash ^= uint32(key[i])
    }
    return hash
}

-- 1. 서브쿼리를 JOIN으로 변환
-- BAD
SELECT * FROM orders
WHERE user_id IN (SELECT id FROM users WHERE status = 'active');

-- GOOD
SELECT o.* FROM orders o
INNER JOIN users u ON o.user_id = u.id
WHERE u.status = 'active';

-- 2. EXISTS vs IN (대량 데이터)
-- GOOD: EXISTS (서브쿼리 결과가 큰 경우)
SELECT * FROM orders o
WHERE EXISTS (
  SELECT 1 FROM users u
  WHERE u.id = o.user_id AND u.status = 'active'
);

-- 3. 페이지네이션 최적화
-- BAD: OFFSET 기반 (깊은 페이지에서 느림)
SELECT * FROM products ORDER BY id LIMIT 20 OFFSET 10000;

-- GOOD: 커서 기반 (일정한 성능)
SELECT * FROM products
WHERE id > 10000
ORDER BY id
LIMIT 20;

-- 4. 집계 쿼리 최적화
-- BAD: COUNT(*)를 자주 호출
SELECT COUNT(*) FROM orders WHERE status = 'pending';

-- GOOD: 대략적인 카운트 사용 (PostgreSQL)
SELECT reltuples::bigint AS estimate
FROM pg_class WHERE relname = 'orders';

-- 5. 파티셔닝
CREATE TABLE orders (
  id BIGSERIAL,
  user_id BIGINT NOT NULL,
  status VARCHAR(20) NOT NULL,
  created_at TIMESTAMP NOT NULL,
  total_amount DECIMAL(10,2)
) PARTITION BY RANGE (created_at);

CREATE TABLE orders_2025_q1 PARTITION OF orders
  FOR VALUES FROM ('2025-01-01') TO ('2025-04-01');
CREATE TABLE orders_2025_q2 PARTITION OF orders
  FOR VALUES FROM ('2025-04-01') TO ('2025-07-01');

# SQLAlchemy - 읽기/쓰기 분리
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

class DatabaseRouter:
    def __init__(self):
        self.writer = create_engine(
            'postgresql://writer:pass@primary:5432/mydb',
            pool_size=10,
            max_overflow=20
        )
        self.readers = [
            create_engine(
                f'postgresql://reader:pass@replica{i}:5432/mydb',
                pool_size=10,
                max_overflow=20
            )
            for i in range(1, 4)  # 3개의 읽기 복제본
        ]
        self._reader_index = 0

    def get_writer_session(self):
        Session = sessionmaker(bind=self.writer)
        return Session()

    def get_reader_session(self):
        # 라운드 로빈으로 읽기 복제본 선택
        reader = self.readers[self._reader_index % len(self.readers)]
        self._reader_index += 1
        Session = sessionmaker(bind=reader)
        return Session()


# 사용 예시
db = DatabaseRouter()

# 쓰기 작업
with db.get_writer_session() as session:
    new_order = Order(user_id=1, total=99.99)
    session.add(new_order)
    session.commit()

# 읽기 작업 (복제본 사용)
with db.get_reader_session() as session:
    orders = session.query(Order).filter_by(status='pending').all()

import redis
import json
from functools import wraps

redis_client = redis.Redis(host='localhost', port=6379, db=0)

class CacheAside:
    """Cache-Aside (Lazy Loading) 패턴 구현"""

    @staticmethod
    def cached(key_prefix, ttl_seconds=300):
        def decorator(func):
            @wraps(func)
            async def wrapper(*args, **kwargs):
                # 캐시 키 생성
                cache_key = f"{key_prefix}:{':'.join(str(a) for a in args)}"

                # 1. 캐시에서 조회
                cached = redis_client.get(cache_key)
                if cached:
                    return json.loads(cached)

                # 2. 캐시 미스 - DB에서 조회
                result = await func(*args, **kwargs)

                # 3. 결과를 캐시에 저장
                if result is not None:
                    redis_client.setex(
                        cache_key,
                        ttl_seconds,
                        json.dumps(result, default=str)
                    )

                return result
            return wrapper
        return decorator

    @staticmethod
    def invalidate(key_pattern):
        """패턴 기반 캐시 무효화"""
        keys = redis_client.keys(key_pattern)
        if keys:
            redis_client.delete(*keys)


# 사용 예시
class ProductService:
    @CacheAside.cached('product', ttl_seconds=600)
    async def get_product(self, product_id):
        return await self.db.products.find_one({"_id": product_id})

    @CacheAside.cached('product:list', ttl_seconds=120)
    async def list_products(self, category, page):
        return await self.db.products.find(
            {"category": category}
        ).skip((page - 1) * 20).limit(20).to_list(20)

    async def update_product(self, product_id, data):
        await self.db.products.update_one(
            {"_id": product_id},
            {"$set": data}
        )
        # 관련 캐시 무효화
        CacheAside.invalidate(f'product:{product_id}')
        CacheAside.invalidate('product:list:*')

class WriteThrough:
    """Write-Through: 캐시와 DB를 동시에 업데이트"""

    async def update(self, key, value, ttl=300):
        # 1. DB에 쓰기
        await self.db.update(key, value)
        # 2. 캐시 업데이트 (DB 쓰기 성공 후)
        redis_client.setex(f"wt:{key}", ttl, json.dumps(value, default=str))

    async def get(self, key):
        # 캐시에서 조회 (항상 최신 데이터)
        cached = redis_client.get(f"wt:{key}")
        if cached:
            return json.loads(cached)
        # 캐시 미스 시 DB 조회 후 캐시 저장
        value = await self.db.get(key)
        if value:
            redis_client.setex(f"wt:{key}", 300, json.dumps(value, default=str))
        return value


class WriteBehind:
    """Write-Behind (Write-Back): 캐시에 먼저 쓰고, 비동기로 DB에 반영"""

    def __init__(self):
        self.write_queue = asyncio.Queue()
        self.batch_size = 100
        self.flush_interval = 5  # 초

    async def update(self, key, value, ttl=300):
        # 1. 캐시에 즉시 쓰기 (빠른 응답)
        redis_client.setex(f"wb:{key}", ttl, json.dumps(value, default=str))
        # 2. 큐에 추가 (비동기 DB 쓰기)
        await self.write_queue.put((key, value))

    async def flush_worker(self):
        """백그라운드 워커: 큐에서 꺼내서 DB에 배치 쓰기"""
        while True:
            batch = []
            try:
                while len(batch) < self.batch_size:
                    item = await asyncio.wait_for(
                        self.write_queue.get(),
                        timeout=self.flush_interval
                    )
                    batch.append(item)
            except asyncio.TimeoutError:
                pass

            if batch:
                try:
                    await self.db.bulk_update(batch)
                except Exception as e:
                    # 실패 시 재시도 큐에 추가
                    for item in batch:
                        await self.write_queue.put(item)
                    await asyncio.sleep(1)

# 다층 TTL 전략
class TieredTTLCache:
    TTL_CONFIG = {
        # 자주 변경되는 데이터
        'user:session': 1800,          # 30분
        'cart:items': 900,             # 15분

        # 주기적으로 변경되는 데이터
        'product:detail': 3600,        # 1시간
        'product:list': 600,           # 10분
        'search:results': 300,         # 5분

        # 거의 변경되지 않는 데이터
        'category:list': 86400,        # 24시간
        'config:settings': 86400,      # 24시간
        'static:content': 604800,      # 7일
    }

    @classmethod
    def get_ttl(cls, key_type):
        return cls.TTL_CONFIG.get(key_type, 300)  # 기본 5분

    @staticmethod
    def stale_while_revalidate(key, ttl, stale_ttl):
        """Stale-While-Revalidate 패턴"""
        cached = redis_client.get(key)
        if cached:
            data = json.loads(cached)
            if data['_cached_at'] + ttl > time.time():
                return data['value'], False  # 신선한 데이터
            if data['_cached_at'] + stale_ttl > time.time():
                return data['value'], True   # 부실하지만 사용 가능
        return None, True  # 캐시 미스

# 캐시 워밍(Pre-warming)
class CacheWarmer:
    async def warm_popular_products(self):
        """인기 상품 캐시 사전 로딩"""
        popular = await self.db.products.find(
            {"popular": True}
        ).limit(100).to_list(100)

        pipe = redis_client.pipeline()
        for product in popular:
            key = f"product:{product['_id']}"
            pipe.setex(key, 3600, json.dumps(product, default=str))
        pipe.execute()

# Celery를 사용한 비동기 태스크 처리
from celery import Celery, chain, group, chord

app = Celery('tasks', broker='redis://localhost:6379/0')

# 설정
app.conf.update(
    task_serializer='json',
    accept_content=['json'],
    result_serializer='json',
    timezone='UTC',
    task_acks_late=True,
    worker_prefetch_multiplier=1,
    task_reject_on_worker_lost=True,
    task_routes={
        'tasks.send_email': {'queue': 'email'},
        'tasks.process_image': {'queue': 'image'},
        'tasks.generate_report': {'queue': 'report'},
    }
)

@app.task(bind=True, max_retries=3, default_retry_delay=60)
def send_email(self, to, subject, body):
    try:
        email_service.send(to=to, subject=subject, body=body)
    except Exception as exc:
        self.retry(exc=exc)

@app.task(bind=True, max_retries=3)
def process_order(self, order_id):
    """주문 처리 파이프라인"""
    try:
        order = Order.objects.get(id=order_id)

        # 체인으로 순차 처리
        workflow = chain(
            validate_inventory.s(order_id),
            process_payment.s(order_id),
            send_confirmation_email.s(order_id),
            update_analytics.s(order_id)
        )
        workflow.apply_async()

    except Exception as exc:
        self.retry(exc=exc, countdown=30)

@app.task
def validate_inventory(result, order_id):
    # 재고 확인
    order = Order.objects.get(id=order_id)
    for item in order.items.all():
        if item.product.stock < item.quantity:
            raise InsufficientStockError(item.product.name)
    return True

@app.task
def bulk_process_orders(order_ids):
    """병렬 배치 처리"""
    job = group(process_order.s(oid) for oid in order_ids)
    result = job.apply_async()
    return result

// Node.js - EventEmitter 기반 비동기 처리
const EventEmitter = require('events');

class OrderEventBus extends EventEmitter {
  constructor() {
    super();
    this.setMaxListeners(20);
  }
}

const orderBus = new OrderEventBus();

// 이벤트 핸들러 등록 (관심사 분리)
orderBus.on('order.created', async (order) => {
  // 재고 업데이트
  await inventoryService.decrementStock(order.items);
});

orderBus.on('order.created', async (order) => {
  // 확인 이메일 발송
  await emailService.sendOrderConfirmation(order);
});

orderBus.on('order.created', async (order) => {
  // 분석 데이터 업데이트
  await analyticsService.trackOrder(order);
});

orderBus.on('order.created', async (order) => {
  // 추천 시스템 업데이트
  await recommendationService.recordPurchase(order.userId, order.items);
});

// 주문 생성 시 이벤트 발행
class OrderService {
  async createOrder(orderData) {
    const order = await this.orderRepo.create(orderData);

    // 동기적으로 필수 작업만 수행
    // 나머지는 이벤트로 비동기 처리
    orderBus.emit('order.created', order);

    return order; // 빠르게 응답
  }
}

# SQLAlchemy 벌크 인서트 비교
import time

# BAD: 하나씩 삽입 (N번의 INSERT)
def insert_one_by_one(session, records):
    start = time.time()
    for record in records:
        session.add(MyModel(**record))
    session.commit()
    print(f"One by one: {time.time() - start:.2f}s")

# GOOD: 벌크 삽입 (1번의 INSERT)
def bulk_insert(session, records):
    start = time.time()
    session.bulk_insert_mappings(MyModel, records)
    session.commit()
    print(f"Bulk insert: {time.time() - start:.2f}s")

# BETTER: execute_values (PostgreSQL, psycopg2)
def execute_values_insert(conn, records):
    start = time.time()
    from psycopg2.extras import execute_values
    cursor = conn.cursor()
    execute_values(
        cursor,
        "INSERT INTO my_table (col1, col2, col3) VALUES %s",
        [(r['col1'], r['col2'], r['col3']) for r in records],
        page_size=1000
    )
    conn.commit()
    print(f"execute_values: {time.time() - start:.2f}s")

# 성능 비교 (10,000건 기준)
# One by one: 12.5s
# Bulk insert: 0.8s
# execute_values: 0.3s

// 외부 API 배치 호출 최적화
class BatchAPIClient {
  constructor(options = {}) {
    this.batchSize = options.batchSize || 50;
    this.concurrency = options.concurrency || 5;
    this.retryAttempts = options.retryAttempts || 3;
    this.delayBetweenBatches = options.delayMs || 100;
  }

  async processBatch(items, processFn) {
    const results = [];
    const errors = [];

    // 아이템을 배치로 분할
    const batches = [];
    for (let i = 0; i < items.length; i += this.batchSize) {
      batches.push(items.slice(i, i + this.batchSize));
    }

    // 동시성 제한하여 배치 처리
    for (let i = 0; i < batches.length; i += this.concurrency) {
      const concurrentBatches = batches.slice(i, i + this.concurrency);

      const batchResults = await Promise.allSettled(
        concurrentBatches.map(batch => this.processWithRetry(batch, processFn))
      );

      for (const result of batchResults) {
        if (result.status === 'fulfilled') {
          results.push(...result.value);
        } else {
          errors.push(result.reason);
        }
      }

      // 배치 간 딜레이 (Rate limiting 방지)
      if (i + this.concurrency < batches.length) {
        await new Promise(r => setTimeout(r, this.delayBetweenBatches));
      }
    }

    return { results, errors, total: items.length, processed: results.length };
  }

  async processWithRetry(batch, processFn, attempt = 1) {
    try {
      return await processFn(batch);
    } catch (error) {
      if (attempt < this.retryAttempts) {
        const delay = Math.pow(2, attempt) * 1000; // 지수 백오프
        await new Promise(r => setTimeout(r, delay));
        return this.processWithRetry(batch, processFn, attempt + 1);
      }
      throw error;
    }
  }
}

// 사용 예시
const client = new BatchAPIClient({ batchSize: 100, concurrency: 3 });

const result = await client.processBatch(userIds, async (batch) => {
  const response = await fetch('/api/users/batch', {
    method: 'POST',
    body: JSON.stringify({ ids: batch }),
    headers: { 'Content-Type': 'application/json' }
  });
  return response.json();
});

// Express.js 압축 설정
const compression = require('compression');

app.use(compression({
  filter: (req, res) => {
    if (req.headers['x-no-compression']) return false;
    return compression.filter(req, res);
  },
  level: 6,              // 압축 레벨 (1-9, 6이 균형점)
  threshold: 1024,       // 1KB 이상만 압축
  memLevel: 8,           // 메모리 사용량 (1-9)
}));

// HTTP/2 서버 설정
const http2 = require('http2');
const fs = require('fs');

const server = http2.createSecureServer({
  key: fs.readFileSync('server.key'),
  cert: fs.readFileSync('server.crt'),
  allowHTTP1: true,
});

server.on('stream', (stream, headers) => {
  const path = headers[':path'];

  // Server Push
  if (path === '/index.html') {
    stream.pushStream({ ':path': '/styles.css' }, (err, pushStream) => {
      if (!err) {
        pushStream.respond({ ':status': 200, 'content-type': 'text/css' });
        pushStream.end(fs.readFileSync('styles.css'));
      }
    });
  }

  stream.respond({
    ':status': 200,
    'content-type': 'text/html',
  });
  stream.end(fs.readFileSync(`.${path}`));
});

# Python requests - 세션 재사용
import requests
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry

# BAD: 매번 새 연결
def fetch_bad(urls):
    results = []
    for url in urls:
        response = requests.get(url)  # 매번 TCP 핸드셰이크
        results.append(response.json())
    return results

# GOOD: 세션 재사용 (Keep-Alive)
def fetch_good(urls):
    session = requests.Session()

    # 재시도 설정
    retry_strategy = Retry(
        total=3,
        backoff_factor=0.5,
        status_forcelist=[500, 502, 503, 504]
    )
    adapter = HTTPAdapter(
        max_retries=retry_strategy,
        pool_connections=10,
        pool_maxsize=20,
        pool_block=False
    )
    session.mount("http://", adapter)
    session.mount("https://", adapter)

    results = []
    for url in urls:
        response = session.get(url)  # 연결 재사용
        results.append(response.json())

    session.close()
    return results

// JSON vs MessagePack vs Protobuf 비교
const msgpack = require('msgpack-lite');

// 테스트 데이터
const data = {
  users: Array.from({ length: 1000 }, (_, i) => ({
    id: i,
    name: `User ${i}`,
    email: `user${i}@example.com`,
    age: 20 + (i % 50),
    active: i % 3 !== 0,
    tags: ['tag1', 'tag2', 'tag3'],
    metadata: { loginCount: i * 10, lastLogin: new Date().toISOString() }
  }))
};

// JSON
console.time('json-serialize');
const jsonStr = JSON.stringify(data);
console.timeEnd('json-serialize');
console.log(`JSON size: ${Buffer.byteLength(jsonStr)} bytes`);

// MessagePack
console.time('msgpack-serialize');
const msgpackBuf = msgpack.encode(data);
console.timeEnd('msgpack-serialize');
console.log(`MessagePack size: ${msgpackBuf.length} bytes`);

// 일반적인 결과:
// JSON size: ~120KB, serialize: ~3ms
// MessagePack size: ~85KB, serialize: ~2ms (약 30% 작음)

// Apache Commons Pool2 기반 객체 풀
import org.apache.commons.pool2.BasePooledObjectFactory;
import org.apache.commons.pool2.PooledObject;
import org.apache.commons.pool2.impl.DefaultPooledObject;
import org.apache.commons.pool2.impl.GenericObjectPool;
import org.apache.commons.pool2.impl.GenericObjectPoolConfig;

public class ExpensiveObjectPool {

    private final GenericObjectPool<ExpensiveObject> pool;

    public ExpensiveObjectPool() {
        GenericObjectPoolConfig<ExpensiveObject> config = new GenericObjectPoolConfig<>();
        config.setMaxTotal(50);
        config.setMaxIdle(20);
        config.setMinIdle(5);
        config.setTestOnBorrow(true);
        config.setTestWhileIdle(true);
        config.setTimeBetweenEvictionRunsMillis(30000);

        pool = new GenericObjectPool<>(new ExpensiveObjectFactory(), config);
    }

    public ExpensiveObject borrow() throws Exception {
        return pool.borrowObject();
    }

    public void returnObject(ExpensiveObject obj) {
        pool.returnObject(obj);
    }

    static class ExpensiveObjectFactory extends BasePooledObjectFactory<ExpensiveObject> {
        @Override
        public ExpensiveObject create() {
            return new ExpensiveObject(); // 비용이 큰 초기화
        }

        @Override
        public PooledObject<ExpensiveObject> wrap(ExpensiveObject obj) {
            return new DefaultPooledObject<>(obj);
        }

        @Override
        public void passivateObject(PooledObject<ExpensiveObject> pooledObj) {
            pooledObj.getObject().reset(); // 풀에 반환 시 상태 초기화
        }

        @Override
        public boolean validateObject(PooledObject<ExpensiveObject> pooledObj) {
            return pooledObj.getObject().isValid();
        }
    }
}

# Prometheus 알림 규칙
groups:
  - name: slo-alerts
    rules:
      # p99 지연시간 SLO 위반
      - alert: HighP99Latency
        expr: |
          histogram_quantile(0.99,
            rate(http_request_duration_seconds_bucket[5m])
          ) > 0.5
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "p99 latency exceeds 500ms"
          description: "p99 latency is at {{ $value }}s for 5 minutes"

      # 에러율 SLO 위반
      - alert: HighErrorRate
        expr: |
          sum(rate(http_requests_total{status=~"5.."}[5m]))
          /
          sum(rate(http_requests_total[5m])) > 0.01
        for: 3m
        labels:
          severity: critical
        annotations:
          summary: "Error rate exceeds 1%"

      # 처리량 급감
      - alert: ThroughputDrop
        expr: |
          sum(rate(http_requests_total[5m]))
          < 0.5 * sum(rate(http_requests_total[5m] offset 1h))
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "Throughput dropped over 50% compared to 1h ago"

      # 커넥션 풀 고갈 임박
      - alert: ConnectionPoolExhaustion
        expr: |
          hikaricp_connections_active
          / hikaricp_connections_max > 0.85
        for: 2m
        labels:
          severity: warning
        annotations:
          summary: "Connection pool utilization above 85%"

      # GC 일시정지 시간 증가
      - alert: HighGCPauseTime
        expr: |
          rate(jvm_gc_pause_seconds_sum[5m])
          / rate(jvm_gc_pause_seconds_count[5m]) > 0.1
        for: 5m
        labels:
          severity: warning
        annotations:
          summary: "Average GC pause time exceeds 100ms"