DataOps-platform/deployment/scripts/deploy_dataops.sh

#!/usr/bin/env bash
#
# DataOps Platform 生产部署脚本
# - 安装 /etc/dataops-platform/dataops.env
# - 创建/更新虚拟环境
# - 配置 Supervisor，通过 run_dataops.sh 加载环境变量并启动 Gunicorn
#

set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

# shellcheck source=dataops-common.sh
source "${SCRIPT_DIR}/dataops-common.sh"

PYTHON_BIN="${PYTHON_BIN:-python3}"
RECREATE_VENV="${RECREATE_VENV:-0}"

check_python() {
  echo_step "检查 Python..."
  if ! command -v "${PYTHON_BIN}" >/dev/null 2>&1; then
    echo_error "${PYTHON_BIN} 未安装"
    exit 1
  fi
  echo_info "Python: $(${PYTHON_BIN} --version 2>&1)"
  if command -v apt-get >/dev/null 2>&1; then
    if ! dpkg -l 2>/dev/null | grep -q python3-venv; then
      echo_warn "安装 python3-venv / python3-pip..."
      apt-get update && apt-get install -y python3-venv python3-pip
    fi
  fi
}

check_supervisor_pkg() {
  echo_step "检查 Supervisor..."
  if ! command -v supervisord >/dev/null 2>&1; then
    if command -v apt-get >/dev/null 2>&1; then
      apt-get update && apt-get install -y supervisor
      systemctl enable supervisor
      systemctl start supervisor
    else
      echo_error "Supervisor 未安装"
      exit 1
    fi
  fi
  echo_info "Supervisor 已就绪"
}

create_directories() {
  echo_step "创建目录..."
  mkdir -p "${APP_DIR}" "${LOG_DIR}" /data/upload /data/archive /var/log/supervisor
  chown -R "${APP_USER}:${APP_GROUP}" "${APP_DIR}" "${LOG_DIR}" /data/upload /data/archive
  echo_info "应用目录: ${APP_DIR}"
  echo_info "日志目录: ${LOG_DIR}"
}

setup_env_file() {
  echo_step "配置环境变量..."
  ensure_env_file "${SCRIPT_DIR}"
  normalize_env_file
  load_env_file
  ensure_env_file_permissions
}

setup_venv() {
  echo_step "配置 Python 虚拟环境..."
  if [[ "${RECREATE_VENV}" == "1" && -d "${VENV_DIR}" ]]; then
    echo_warn "重建虚拟环境: ${VENV_DIR}"
    rm -rf "${VENV_DIR}"
  fi
  if [[ ! -d "${VENV_DIR}" ]]; then
    sudo -u "${APP_USER}" "${PYTHON_BIN}" -m venv "${VENV_DIR}"
  fi
  "${VENV_DIR}/bin/pip" install --upgrade pip
  if [[ ! -f "${APP_DIR}/requirements.txt" ]]; then
    echo_error "缺少 ${APP_DIR}/requirements.txt"
    exit 1
  fi
  "${VENV_DIR}/bin/pip" install -r "${APP_DIR}/requirements.txt"
  chown -R "${APP_USER}:${APP_GROUP}" "${VENV_DIR}"
  echo_info "依赖安装完成"
}

verify_app() {
  echo_step "验证应用..."
  load_env_file
  sudo -u "${APP_USER}" env APP_ENV_FILE="${ENV_FILE}" FLASK_ENV=production \
    bash -c "set -a; source <(sed 's/\\r$//' '${ENV_FILE}' | sed '1s/^\\xEF\\xBB\\xBF//'); set +a; cd '${APP_DIR}'; '${VENV_DIR}/bin/python' -c \"
from app import create_app
app = create_app()
print('Flask 应用创建成功')
print('蓝图:', list(app.blueprints.keys()))
\"" || {
    echo_error "应用验证失败，请检查 ${ENV_FILE} 中的数据库/Neo4j 等配置"
    exit 1
  }
  echo_info "应用验证通过"
}

sync_ops_scripts() {
  echo_step "同步运维脚本..."
  mkdir -p "${APP_DIR}/scripts"
  install -m 755 "${SCRIPT_DIR}/run_dataops.sh" "${RUN_SCRIPT}"
  for script in deploy_dataops.sh start_dataops.sh stop_dataops.sh restart_dataops.sh dataops-common.sh; do
    if [[ -f "${SCRIPT_DIR}/${script}" ]]; then
      install -m 755 "${SCRIPT_DIR}/${script}" "${SCRIPTS_DIR}/${script}"
    fi
  done
  chown -R "${APP_USER}:${APP_GROUP}" "${SCRIPTS_DIR}"
  normalize_shell_scripts "${SCRIPTS_DIR}"
  ensure_run_script
}

start_application() {
  echo_step "启动应用..."
  supervisorctl restart "${APP_NAME}" 2>/dev/null || supervisorctl start "${APP_NAME}"
  sleep 8
  local status
  status="$(supervisorctl status "${APP_NAME}" | awk '{print $2}')"
  if [[ "${status}" == "RUNNING" ]]; then
    echo_info "应用已运行"
    supervisorctl status "${APP_NAME}"
  else
    echo_error "应用启动失败，状态: ${status}"
    echo_info "查看日志: tail -f ${SUPERVISOR_LOG}"
    exit 1
  fi
}

show_summary() {
  local app_port
  app_port="$(resolve_listen_port)"
  echo ""
  echo "=========================================="
  echo -e "${GREEN}  部署完成${NC}"
  echo "=========================================="
  echo "  应用目录: ${APP_DIR}"
  echo "  环境变量: ${ENV_FILE}"
  echo "  监听地址: ${LISTEN_HOST}:${app_port}"
  echo ""
  echo "  启动: sudo ${APP_DIR}/scripts/start_dataops.sh"
  echo "  停止: sudo ${APP_DIR}/scripts/stop_dataops.sh"
  echo "  重启: sudo ${APP_DIR}/scripts/restart_dataops.sh"
  echo "  健康: curl http://127.0.0.1:${app_port}/api/system/health"
  echo ""
}

main() {
  echo "=========================================="
  echo "  DataOps Platform 部署脚本"
  echo "=========================================="
  require_root
  check_python
  check_supervisor_pkg
  create_directories
  setup_env_file
  ensure_wsgi "${SCRIPT_DIR}"
  setup_venv
  verify_app
  sync_ops_scripts
  load_env_file
  configure_supervisor
  start_application
  if ! health_check; then
    echo_error "部署完成但健康检查失败，请根据上方诊断信息排查"
    exit 1
  fi
  show_summary
}

main "$@"