order_site/docker-compose.yml

x-json-log-rotation: &json-log-rotation
  driver: json-file
  options:
    max-size: "20m"
    max-file: "5"

services:
  postgres:
    image: docker.m.daocloud.io/library/postgres:16-bookworm
    # 开启 pg_stat_statements，便于定位生产环境累计最耗时的 SQL。
    # 内存类参数默认按 4 核 8G 且与其他服务共用的服务器设置（见 .env.server.example）。
    command:
      [
        "postgres",
        "-c",
        "shared_preload_libraries=pg_stat_statements",
        "-c",
        "pg_stat_statements.track=all",
        "-c",
        "log_min_duration_statement=${POSTGRES_LOG_MIN_DURATION_STATEMENT_MS:-500}",
        "-c",
        "shared_buffers=${POSTGRES_SHARED_BUFFERS:-384MB}",
        "-c",
        "effective_cache_size=${POSTGRES_EFFECTIVE_CACHE_SIZE:-4GB}",
        "-c",
        "work_mem=${POSTGRES_WORK_MEM:-16MB}",
        "-c",
        "maintenance_work_mem=${POSTGRES_MAINTENANCE_WORK_MEM:-256MB}",
        # 并行度 1：生产实测单条大查询一旦开 2 个 parallel worker，就需要
        # 「leader + 2 worker = 3 核」，而容器配额只有 2 核 → 必然撞配额被节流
        # （累计 nr_throttled 94,197 次 / throttled_usec 9,337s）。重查询已优化到
        # 10ms 量级，用不到并行。
        "-c",
        "max_parallel_workers_per_gather=${POSTGRES_MAX_PARALLEL_WORKERS_PER_GATHER:-1}",
        # 检查点：默认 5 分钟一次 + max_wal_size 1GB，生产 43 天累计 21,778 次检查点，
        # sar 里就是每 20–40 分钟一轮写突发（wtps 130、8MB/s）。放宽到 15 分钟 / 4GB。
        "-c",
        "checkpoint_timeout=${POSTGRES_CHECKPOINT_TIMEOUT:-900}",
        "-c",
        "max_wal_size=${POSTGRES_MAX_WAL_SIZE:-4GB}",
        # 自动清理观测：默认 600000（10 分钟）等于什么都看不到。
        # 这是 sighup 级参数，不能 ALTER DATABASE SET，只能走启动参数。
        "-c",
        "log_autovacuum_min_duration=${POSTGRES_LOG_AUTOVACUUM_MIN_DURATION_MS:-5000}",
      ]
    restart: unless-stopped
    # 让 PostgreSQL 优雅关闭有足够时间完成检查点；无限制时 OOM 可能拖垮整机。
    mem_limit: ${POSTGRES_MEM_LIMIT:-1.5g}
    # 3 核：这台机器 4 核，postgres 是最大单一消耗者（22.6h 烧 12.2 CPU·h、均 0.54 核），
    # 却被 2 核配额卡住，累计 2.6 小时处于被节流状态。同机其它容器实测只用 0.2–0.4 核，
    # 且几乎不节流，所以这里让出 1 核给数据库的突发。
    cpus: ${POSTGRES_CPUS:-3}
    shm_size: 512mb
    stop_grace_period: 60s
    logging: *json-log-rotation
    environment:
      TZ: ${TZ:-Asia/Shanghai}
      POSTGRES_DB: ${POSTGRES_DB:-order_site}
      POSTGRES_USER: ${POSTGRES_USER:-postgres}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-postgres}
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "pg_isready -U ${POSTGRES_USER:-postgres} -d ${POSTGRES_DB:-order_site}",
        ]
      interval: 5s
      timeout: 5s
      retries: 20
    volumes:
      - postgres_data:/var/lib/postgresql/data

  backend:
    build:
      context: .
      dockerfile: deploy/docker/backend.Dockerfile
      args:
        # 默认阿里云公共源；阿里云 ECS 内网可设 DEBIAN_MIRROR=mirrors.cloud.aliyuncs.com
        DEBIAN_MIRROR: ${DEBIAN_MIRROR:-mirrors.aliyun.com}
        NPM_REGISTRY: ${NPM_REGISTRY:-https://registry.npmmirror.com}
    restart: unless-stopped
    # 30s 让后端在 SIGTERM 后完成优雅关闭（HTTP 连接收尾 + 数据库连接池释放）。
    # 内存限额 768m → 1g：生产实测十天从重启后的 ~194MB 涨到 754MB(98%)，
    # 贴着 cgroup 上限时 V8 会持续 GC 抖动、接口整体变慢（且未必 OOM，所以日志里看不到崩溃）。
    # 服务器 .env 里若仍写着 BACKEND_MEM_LIMIT=768m，会覆盖这里的默认值，需一并调整。
    mem_limit: ${BACKEND_MEM_LIMIT:-1g}
    cpus: ${BACKEND_CPUS:-1.5}
    stop_grace_period: 30s
    env_file:
      - .env
    depends_on:
      postgres:
        condition: service_healthy
    healthcheck:
      test:
        ["CMD-SHELL", "curl -fsS http://127.0.0.1:3000/health/ready || exit 1"]
      interval: 5s
      timeout: 5s
      retries: 20
      start_period: 20s
    environment:
      TZ: ${TZ:-Asia/Shanghai}
      PORT: ${BACKEND_PORT:-3000}
      # V8 老生代上限留出余量（限额 1g → 堆 768m）：让 GC 在撞上 cgroup 墙之前主动回收，
      # 避免"贴顶后才疯狂 GC"的抖动。native/external 内存与堆外缓冲还需要剩余那部分。
      NODE_OPTIONS: ${BACKEND_NODE_OPTIONS:---max-old-space-size=768}
      HTTP_TRUST_PROXY: ${HTTP_TRUST_PROXY:-1}
      DATABASE_URL: ${DATABASE_URL:-postgres://postgres:postgres@postgres:5432/order_site}
      DATABASE_SSL: ${DATABASE_SSL:-false}
      DATABASE_MAX_CONNECTIONS: ${DATABASE_MAX_CONNECTIONS:-8}
      DATABASE_IDLE_TIMEOUT_MS: ${DATABASE_IDLE_TIMEOUT_MS:-30000}
      DATABASE_CONNECTION_TIMEOUT_MS: ${DATABASE_CONNECTION_TIMEOUT_MS:-5000}
      DATABASE_STATEMENT_TIMEOUT_MS: ${DATABASE_STATEMENT_TIMEOUT_MS:-15000}
      DATA_ROOT: /app/data
      LOG_RETENTION_DAYS: ${LOG_RETENTION_DAYS:-30}
      LOG_MAX_FILE_SIZE_MB: ${LOG_MAX_FILE_SIZE_MB:-20}
      LOG_MAX_FILES: ${LOG_MAX_FILES:-5}
      CLAIM_BASE_URL: ${CLAIM_BASE_URL:-http://localhost/#/claim}
      # 生产图片存储只走阿里云 OSS；MinIO 仅保留在本地开发编排（docker-compose.dev.yml）。
      STORAGE_MODE: ${STORAGE_MODE:-oss}
      STORAGE_ENDPOINT: ${STORAGE_ENDPOINT:-http://minio:9000}
      STORAGE_BUCKET: ${STORAGE_BUCKET:-order-site}
      STORAGE_ACCESS_KEY_ID: ${STORAGE_ACCESS_KEY_ID:-minioadmin}
      STORAGE_SECRET_ACCESS_KEY: ${STORAGE_SECRET_ACCESS_KEY:-minioadmin}
      STORAGE_REGION: ${STORAGE_REGION:-us-east-1}
      STORAGE_MAX_UPLOAD_SIZE_MB: ${STORAGE_MAX_UPLOAD_SIZE_MB:-10}
      STORAGE_OSS_ENDPOINT: ${STORAGE_OSS_ENDPOINT:-}
      STORAGE_OSS_BUCKET: ${STORAGE_OSS_BUCKET:-}
      STORAGE_OSS_ACCESS_KEY_ID: ${STORAGE_OSS_ACCESS_KEY_ID:-}
      STORAGE_OSS_SECRET_ACCESS_KEY: ${STORAGE_OSS_SECRET_ACCESS_KEY:-}
      STORAGE_OSS_REGION: ${STORAGE_OSS_REGION:-oss-cn-hangzhou}
      # ADMIN_SESSION_SECRET / ADMIN_DEFAULT_USERS_JSON 仅走 env_file。
      # 勿在此处用 ${ADMIN_DEFAULT_USERS_JSON} 再注入：JSON 会被 YAML 解析破坏，导致生产校验失败。
      KUASHOU_INDUSTRY_RATE_LIMIT_MAX: ${KUASHOU_INDUSTRY_RATE_LIMIT_MAX:-120}
    logging: *json-log-rotation
    volumes:
      - ./apps/backend/data:/app/data

  web:
    build:
      context: .
      dockerfile: deploy/docker/frontend.Dockerfile
      args:
        APP_DOMAIN: ${APP_DOMAIN:-localhost}
        VITE_WORKER_SITE_URL: ${WORKER_SITE_ADDR:-https://worker.khhao.com}
        ALPINE_MIRROR: ${ALPINE_MIRROR:-mirrors.aliyun.com}
        NPM_REGISTRY: ${NPM_REGISTRY:-https://registry.npmmirror.com}
    restart: unless-stopped
    mem_limit: ${WEB_MEM_LIMIT:-256m}
    cpus: ${WEB_CPUS:-0.5}
    stop_grace_period: 20s
    logging: *json-log-rotation
    depends_on:
      backend:
        condition: service_healthy
    ports:
      - "80:80"
      - "443:443"
    environment:
      TZ: ${TZ:-Asia/Shanghai}
      APP_DOMAIN: ${APP_DOMAIN:-localhost}
      CADDY_SITE_ADDR: ${CADDY_SITE_ADDR:-http://localhost}
      WORKER_SITE_ADDR: ${WORKER_SITE_ADDR:-https://worker.khhao.com}
    volumes:
      - caddy_data:/data
      - caddy_config:/config

volumes:
  postgres_data:
  caddy_data:
  caddy_config: