eth_hedge_sim/deploy/lib/common.sh

#!/usr/bin/env bash
# deploy/lib/common.sh — 比特骆驼自动化对冲系统 部署公共函数
# 目标系统: Ubuntu 22.04 LTS
set -e
set -u
if [ -n "${BASH_VERSION:-}" ]; then
  set -o pipefail
fi

INSTALL_ROOT="${INSTALL_ROOT:-/opt/eth_hedge_sim}"
GIT_URL="${GIT_URL:-https://git.bz121.com/dekun/eth_hedge_sim.git}"
GIT_BRANCH="${GIT_BRANCH:-main}"
BACKUP_ROOT="${BACKUP_ROOT:-/root/backups/eth_hedge_sim}"
TZ_NAME="${EHS_TZ:-Asia/Shanghai}"
NODE_MAJOR="${NODE_MAJOR:-20}"
PM2_APP_NAME="${PM2_APP_NAME:-eth-hedge-api}"
API_PORT="${API_PORT:-5155}"

LIB_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
DEPLOY_DIR="$(cd "${LIB_DIR}/.." && pwd)"
REPO_ROOT="$(cd "${DEPLOY_DIR}/.." && pwd)"

log() { printf '[%s] %s\n' "$(TZ="${TZ_NAME}" date '+%Y-%m-%d %H:%M:%S')" "$*"; }
step() { echo ""; log "==> $*"; }

die() {
  echo "错误: $*" >&2
  exit 1
}

require_root() {
  if [[ "$(id -u)" -ne 0 ]]; then
    die "请使用 root 执行(推荐: sudo -i 后运行)"
  fi
}

# 打印当前占用 apt 的进程，便于排查
_apt_lock_holders() {
  ps -eo pid,cmd 2>/dev/null | grep -E '[u]nattended-upgr|[a]pt-get|[a]pt |[d]pkg ' | head -n 8 || true
  if command -v fuser >/dev/null 2>&1; then
    fuser -v /var/lib/dpkg/lock-frontend /var/lib/dpkg/lock /var/lib/apt/lists/lock 2>&1 | head -n 12 || true
  fi
}

# 部署期间临时停自动升级（重启后常立刻再跑，占锁很久）
_stop_auto_apt_for_deploy() {
  if [[ "${APT_STOP_AUTO:-1}" != "1" ]]; then
    return 0
  fi
  if command -v systemctl >/dev/null 2>&1; then
    log "临时停止 unattended-upgrades / apt-daily，避免占锁…"
    systemctl stop unattended-upgrades.service 2>/dev/null || true
    systemctl stop apt-daily.service apt-daily-upgrade.service 2>/dev/null || true
    systemctl kill --kill-who=all unattended-upgrades.service 2>/dev/null || true
  fi
  # 若仍卡死且 APT_FORCE_UNLOCK=1，才强杀（默认不杀，避免打断半截升级）
  if [[ "${APT_FORCE_UNLOCK:-0}" == "1" ]]; then
    log "APT_FORCE_UNLOCK=1：结束残留 apt/dpkg 进程…"
    pkill -9 -x unattended-upgr 2>/dev/null || true
    pkill -9 -x apt-get 2>/dev/null || true
    pkill -9 -x apt 2>/dev/null || true
    pkill -9 -x dpkg 2>/dev/null || true
    sleep 2
    dpkg --configure -a 2>/dev/null || true
  fi
}

# 等待 apt/dpkg 锁（常见：unattended-upgrades 正在跑）
wait_for_apt_lock() {
  local max_wait="${1:-600}"
  local waited=0
  local lock_frontend="/var/lib/dpkg/lock-frontend"
  local lock_dpkg="/var/lib/dpkg/lock"
  local lock_lists="/var/lib/apt/lists/lock"
  local tried_stop=0

  if ! command -v apt-get >/dev/null 2>&1; then
    return 0
  fi

  while true; do
    local busy=0
    if pgrep -x unattended-upgr >/dev/null 2>&1 \
      || pgrep -x apt-get >/dev/null 2>&1 \
      || pgrep -x apt >/dev/null 2>&1 \
      || pgrep -x dpkg >/dev/null 2>&1; then
      busy=1
    fi
    # fuser 在有进程占用锁文件时返回 0
    if command -v fuser >/dev/null 2>&1; then
      if fuser "${lock_frontend}" >/dev/null 2>&1 \
        || fuser "${lock_dpkg}" >/dev/null 2>&1 \
        || fuser "${lock_lists}" >/dev/null 2>&1; then
        busy=1
      fi
    fi
    if [[ "${busy}" -eq 0 ]]; then
      if [[ "${waited}" -gt 0 ]]; then
        log "apt 锁已释放，继续安装"
      fi
      return 0
    fi
    if [[ "${waited}" -eq 0 ]]; then
      log "检测到 apt/dpkg 正被占用（多为重启后的 unattended-upgrades），等待释放…"
      log "占用进程："
      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
      log "可另开终端: ps aux | grep -E 'unattended|apt|dpkg'"
    elif [[ "${tried_stop}" -eq 0 && "${waited}" -ge 15 ]]; then
      # 等约 15s 仍占用：停自动升级服务（不杀半截 dpkg，除非 APT_FORCE_UNLOCK=1）
      tried_stop=1
      _stop_auto_apt_for_deploy
    elif [[ $((waited % 60)) -eq 0 ]]; then
      log "仍在等待 apt 锁…已等 ${waited}s / ${max_wait}s"
      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
    fi
    if [[ "${waited}" -ge "${max_wait}" ]]; then
      log "超时前占用进程："
      _apt_lock_holders | while IFS= read -r line; do log "  ${line}"; done
      die "等待 apt 锁超时(${max_wait}s)。可手动: systemctl stop unattended-upgrades; 或 APT_FORCE_UNLOCK=1 bash manage.sh"
    fi
    sleep 5
    waited=$((waited + 5))
  done
}

apt_update() {
  wait_for_apt_lock
  apt-get update -qq
}

apt_install() {
  wait_for_apt_lock
  apt-get install -y "$@"
}

detect_server_ip() {
  local ip=""
  if command -v hostname >/dev/null 2>&1; then
    ip="$(hostname -I 2>/dev/null | awk '{print $1}')"
  fi
  if [[ -z "${ip}" ]]; then
    ip="127.0.0.1"
  fi
  echo "${ip}"
}

# curl|bash 时 stdin 可能是管道，交互从 /dev/tty 读
cm_read() {
  local __var="$1"
  local __prompt="${2:-}"
  local __line=""
  if [[ -n "${__prompt}" ]]; then
    printf '%s' "${__prompt}" >/dev/tty 2>/dev/null || printf '%s' "${__prompt}"
  fi
  if [[ -r /dev/tty ]]; then
    IFS= read -r __line </dev/tty || true
  else
    IFS= read -r __line || true
  fi
  printf -v "${__var}" '%s' "${__line}"
}

confirm_yes() {
  local msg="$1"
  local ans=""
  cm_read ans "${msg} [y/N] "
  [[ "${ans}" == [yY] || "${ans}" == [yY][eE][sS] ]]
}

repo_ready() {
  [[ -f "${1}/deploy/manage.sh" && -f "${1}/deploy/pull_and_restart.sh" ]]
}

resolve_repo_root() {
  if repo_ready "${INSTALL_ROOT}"; then
    echo "${INSTALL_ROOT}"
    return 0
  fi
  if [[ -n "${REPO_ROOT:-}" ]] && repo_ready "${REPO_ROOT}"; then
    echo "${REPO_ROOT}"
    return 0
  fi
  return 1
}

# ---------- 运行时检测（有则跳过，无则安装） ----------

python_runtime_ok() {
  command -v python3 >/dev/null 2>&1 || return 1
  # Ubuntu 常有 python3 且 import venv 成功，但缺 python3.X-venv 时 ensurepip 不可用，
  # python3 -m venv 会失败。必须以 ensurepip 为准。
  python3 -c "import ensurepip" >/dev/null 2>&1 || return 1
  return 0
}

node_runtime_ok() {
  command -v node >/dev/null 2>&1 || return 1
  command -v npm >/dev/null 2>&1 || return 1
  return 0
}

pm2_runtime_ok() {
  command -v pm2 >/dev/null 2>&1
}

print_runtime_probe() {
  echo "环境检测:"
  if python_runtime_ok; then
    echo "  [OK] Python3 + venv       →  $(python3 --version 2>&1)"
  else
    echo "  [缺] Python3 / venv       →  将自动安装"
  fi
  if node_runtime_ok; then
    echo "  [OK] Node.js + npm         →  $(node --version 2>&1) / npm $(npm --version 2>&1)"
  else
    echo "  [缺] Node.js / npm         →  将自动安装 Node ${NODE_MAJOR}"
  fi
  if pm2_runtime_ok; then
    echo "  [OK] PM2                   →  $(pm2 -v 2>&1)"
  else
    echo "  [缺] PM2                   →  将自动安装"
  fi
}

install_python_runtime() {
  step "安装 Python3 环境 (python3 / venv / pip / ensurepip)"
  export DEBIAN_FRONTEND=noninteractive
  if ! command -v apt-get >/dev/null 2>&1; then
    die "非 apt 系统，请手动安装 python3 python3-venv python3-pip"
  fi
  apt_update
  apt_install python3 python3-venv python3-pip
  # Debian/Ubuntu：版本化包（如 python3.10-venv）才真正提供 ensurepip
  local pyver=""
  pyver="$(python3 -c 'import sys; print(f"{sys.version_info.major}.{sys.version_info.minor}")' 2>/dev/null || true)"
  if [[ -n "${pyver}" ]]; then
    log "安装 python${pyver}-venv（提供 ensurepip）…"
    apt_install "python${pyver}-venv" || true
  fi
  if ! python_runtime_ok; then
    die "Python 环境安装后仍不可用（缺 ensurepip）。请检查: apt install python3-venv python${pyver:-3}-venv"
  fi
  log "Python 已就绪: $(python3 --version 2>&1)"
}

install_node_runtime() {
  step "安装 Node.js ${NODE_MAJOR}"
  export DEBIAN_FRONTEND=noninteractive
  if ! command -v apt-get >/dev/null 2>&1; then
    die "非 apt 系统，请手动安装 Node.js ${NODE_MAJOR}+"
  fi
  apt_update
  apt_install ca-certificates curl
  wait_for_apt_lock
  curl -fsSL "https://deb.nodesource.com/setup_${NODE_MAJOR}.x" | bash -
  apt_install nodejs
  if ! node_runtime_ok; then
    die "Node.js 安装后仍不可用"
  fi
  log "Node 已就绪: $(node --version 2>&1)"
}

install_pm2_runtime() {
  step "安装 PM2"
  if ! command -v npm >/dev/null 2>&1; then
    die "安装 PM2 需要先有 npm"
  fi
  npm install -g pm2
  if ! pm2_runtime_ok; then
    die "PM2 安装后仍不可用"
  fi
  log "PM2 已就绪: $(pm2 -v 2>&1)"
}

ensure_base_packages() {
  local need=()
  command -v git >/dev/null 2>&1 || need+=(git)
  command -v curl >/dev/null 2>&1 || need+=(curl)
  if [[ ${#need[@]} -eq 0 ]]; then
    log "基础包 git/curl 已存在，跳过"
    return 0
  fi
  if ! command -v apt-get >/dev/null 2>&1; then
    die "缺少依赖: ${need[*]}，且无 apt-get"
  fi
  step "安装基础包: ${need[*]}"
  export DEBIAN_FRONTEND=noninteractive
  apt_update
  apt_install ca-certificates "${need[@]}"
}

ensure_system_deps() {
  step "环境检测 (Python / Node / PM2)"
  print_runtime_probe
  ensure_base_packages

  if python_runtime_ok; then
    log "Python 环境已存在，跳过安装"
  else
    install_python_runtime
  fi

  if node_runtime_ok; then
    log "Node.js 环境已存在，跳过安装"
  else
    install_node_runtime
  fi

  if pm2_runtime_ok; then
    log "PM2 已存在，跳过安装"
  else
    install_pm2_runtime
  fi

  step "环境确认"
  print_runtime_probe
}

pm2_stop_project() {
  if command -v pm2 >/dev/null 2>&1; then
    pm2 stop "${PM2_APP_NAME}" 2>/dev/null || true
  fi
}

pm2_delete_project() {
  if command -v pm2 >/dev/null 2>&1; then
    pm2 delete "${PM2_APP_NAME}" 2>/dev/null || true
    pm2 save 2>/dev/null || true
  fi
}

pm2_save_startup() {
  if command -v pm2 >/dev/null 2>&1; then
    pm2 save 2>/dev/null || true
    pm2 startup systemd -u root --hp /root 2>/dev/null || true
  fi
}

verify_health() {
  local port="${1:-${API_PORT}}"
  local url="http://127.0.0.1:${port}/health"
  step "健康检查 ${url}"
  local i
  for i in 1 2 3 4 5 6 7 8 9 10; do
    if curl -fsS "${url}" >/dev/null 2>&1; then
      log "health OK"
      curl -fsS "${url}" || true
      echo ""
      return 0
    fi
    sleep 1
  done
  log "警告: health 暂未就绪，请稍后检查 pm2 logs ${PM2_APP_NAME}"
  return 1
}

print_post_install_guide() {
  local ip
  ip="$(detect_server_ip)"
  echo ""
  echo "══════════════════════════════════════"
  echo " 比特骆驼自动化对冲系统 部署完成"
  echo " 目录: ${INSTALL_ROOT}"
  echo " 本机: http://${ip}:${API_PORT}/health"
  echo " 进程: pm2 status  →  ${PM2_APP_NAME}"
  echo " 管理: bash ${INSTALL_ROOT}/deploy/manage.sh"
  echo " 配置: ${INSTALL_ROOT}/.env"
  echo "══════════════════════════════════════"
  echo ""
}