DataOps-platform/scripts/run_dataops.sh

#!/usr/bin/env bash
set -euo pipefail

APP_DIR="${APP_DIR:-/opt/dataops-platform}"
ENV_FILE="${APP_ENV_FILE:-/etc/dataops-platform/dataops.env}"
LOG_DIR="${APP_DIR}/logs"
VENV_DIR="${APP_DIR}/venv"

mkdir -p "${LOG_DIR}"

if [[ -f "${ENV_FILE}" ]]; then
  if [[ ! -r "${ENV_FILE}" ]]; then
    echo "无法读取环境变量文件: ${ENV_FILE} (用户: $(id -un))" >&2
    echo "请执行: sudo chown root:ubuntu ${ENV_FILE} && sudo chmod 640 ${ENV_FILE}" >&2
    exit 1
  fi
  set -a
  # shellcheck disable=SC1090
  source <(sed 's/\r$//' "${ENV_FILE}" | sed '1s/^\xEF\xBB\xBF//')
  set +a
fi

if [[ -z "${DEEPSEEK_API_KEY:-}" && -z "${LLM_API_KEY:-}" ]]; then
  echo "[run_dataops] WARN: DEEPSEEK_API_KEY 未从 ${ENV_FILE} 加载" >&2
fi

export FLASK_ENV="${FLASK_ENV:-production}"
export APP_ENV_FILE="${ENV_FILE}"
export APP_DIR="${APP_DIR:-/opt/dataops-platform}"
export PYTHONPATH="${APP_DIR}${PYTHONPATH:+:${PYTHONPATH}}"

LISTEN_HOST="${LISTEN_HOST:-0.0.0.0}"
LISTEN_PORT="${LISTEN_PORT:-5500}"
GUNICORN_WORKERS="${GUNICORN_WORKERS:-4}"
GUNICORN_TIMEOUT="${GUNICORN_TIMEOUT:-120}"

# 与 Flask ProductionConfig.PORT 保持一致
export PORT="${PORT:-${LISTEN_PORT}}"

# 统一写入绝对路径，避免相对路径落到 APP_DIR 根目录
export LOG_DIR="${LOG_DIR:-${APP_DIR}/logs}"
mkdir -p "${LOG_DIR}"
if [[ -z "${LOG_FILE:-}" || "${LOG_FILE}" != /* ]]; then
  log_filename="${LOG_FILE:-flask_production.log}"
  log_filename="${log_filename##*/}"
  export LOG_FILE="${LOG_DIR}/${log_filename}"
fi

cd "${APP_DIR}"

if [[ ! -x "${VENV_DIR}/bin/gunicorn" ]]; then
  echo "gunicorn 未安装: ${VENV_DIR}/bin/gunicorn" >&2
  exit 1
fi

if [[ ! -f "${APP_DIR}/alembic.ini" || ! -d "${APP_DIR}/migrations" ]]; then
  echo "缺少数据库迁移文件: ${APP_DIR}/alembic.ini 或 migrations/" >&2
  exit 1
fi

echo "[run_dataops] applying database migrations" >&2
role_init_url="${DB_ROLE_INIT_DATABASE_URL:?DB_ROLE_INIT_DATABASE_URL is required for one-shot role bootstrap}"
migration_url="${MIGRATION_DATABASE_URL:?MIGRATION_DATABASE_URL is required}"
runtime_url="${DATABASE_URL:?DATABASE_URL runtime connection is required}"
if [[ "${role_init_url}" == "${runtime_url}" || "${migration_url}" == "${runtime_url}" ]]; then
  echo "role-init/migration and runtime URLs must use different database identities" >&2
  exit 1
fi
DB_ROLE_INIT_DATABASE_URL="${role_init_url}" \
  "${VENV_DIR}/bin/python" -m app.core.edge_gateway.runtime_roles
MIGRATION_DATABASE_URL="${migration_url}" \
  "${VENV_DIR}/bin/alembic" -c "${APP_DIR}/alembic.ini" upgrade head
unset DB_ROLE_INIT_DATABASE_URL MIGRATION_DATABASE_URL SQLALCHEMY_DATABASE_URI
unset role_init_url migration_url

echo "[run_dataops] binding ${LISTEN_HOST}:${LISTEN_PORT} workers=${GUNICORN_WORKERS} log=${LOG_FILE}" >&2

if [[ ! -f "${APP_DIR}/gunicorn_config.py" ]]; then
  echo "缺少 gunicorn 配置: ${APP_DIR}/gunicorn_config.py" >&2
  exit 1
fi

# 启动前验证 Flask 应用能正常加载（失败时错误写入 supervisor 日志）
if ! "${VENV_DIR}/bin/python" -c "from app import create_app; create_app(); print('app bootstrap ok')" >&2; then
  echo "Flask 应用预检失败，请检查 ${ENV_FILE} 与依赖配置" >&2
  exit 1
fi

exec "${VENV_DIR}/bin/gunicorn" \
  --config "${APP_DIR}/gunicorn_config.py" \
  --bind "${LISTEN_HOST}:${LISTEN_PORT}" \
  --workers "${GUNICORN_WORKERS}" \
  --timeout "${GUNICORN_TIMEOUT}" \
  --access-logfile "${LOG_DIR}/gunicorn_access.log" \
  --error-logfile "${LOG_DIR}/gunicorn_error.log" \
  --capture-output \
  --enable-stdio-inheritance \
  --log-level info \
  wsgi:application