feat(quality): 第一阶段质量保障 - 数据库安全闸门 + 生产回滚脚本 + E2E修复

1. 数据库迁移安全检查 (scripts/check_migration_safety.py)
   - 检测 DROP TABLE / DROP COLUMN 等高风险破坏性变更
   - 检测列类型变更、NOT NULL新增、重命名等中风险操作
   - 只检查 upgrade 函数，忽略 downgrade 中的正常回滚操作
   - 已加入 CI validate 闸门（--allow-medium-risk 模式）

2. 生产一键回滚脚本 (scripts/rollback_production.sh)
   - 支持指定版本镜像回滚
   - 可选数据库版本回滚
   - 健康检查确认服务正常
   - 二次确认防误操作

3. E2E CI 修复
   - staging-e2e checkout 从 actions/checkout@v4 改为自定义脚本（修复Gitea兼容性）
   - 缩小E2E范围到核心冒烟用例：登录/上传/生成
   - 先保证稳定，再逐步扩大覆盖

.gitea/workflows/ci-cd.yml

@@ -128,6 +128,12 @@ jobs:
           grep -q "Running upgrade" /tmp/alembic-upgrade.sql
           python3 scripts/check_schema_metadata.py
 
+      - name: Check migration safety
+        shell: sh
+        run: |
+          set -eu
+          python3 scripts/check_migration_safety.py --allow-medium-risk
+
       - name: Run unit tests
         shell: sh
         env:
@@ -404,8 +410,28 @@ jobs:
     needs: deploy-staging
 
     steps:
-    - name: Checkout
-      uses: actions/checkout@v4
+    - name: Checkout code
+      shell: sh
+      env:
+        GITHUB_TOKEN: ${{ github.token }}
+      run: |
+        set -eu
+        archive_url="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}/archive/${GITHUB_SHA}.tar.gz"
+        for i in 1 2 3 4 5; do
+          if wget --header="Authorization: token ${GITHUB_TOKEN}" -O /tmp/repo.tar.gz "$archive_url" 2>&1; then
+            break
+          fi
+          if [ "$i" -lt 5 ]; then
+            wait=$((2 ** i))
+            echo "Checkout failed (attempt $i/5), retrying in ${wait}s..."
+            sleep "$wait"
+          else
+            echo "Checkout failed after 5 attempts"
+            exit 1
+          fi
+        done
+        tar -xzf /tmp/repo.tar.gz --strip-components=1 -C .
+        rm -f /tmp/repo.tar.gz
 
     - name: Run Playwright E2E on staging
       shell: sh
@@ -418,7 +444,7 @@ jobs:
           -v "$PWD:/workspace" \
           -w /workspace/apps/web \
           mcr.microsoft.com/playwright:v1.45.0-jammy \
-          sh -lc "npm ci && npx playwright test --reporter=line --project=chromium"
+          sh -lc "npm ci && npx playwright test --reporter=line --project=chromium e2e/auth.spec.ts e2e/core-upload.spec.ts e2e/core-generation.spec.ts"
 
   build-production-runtime-images:
     name: Build Production Runtime Images

scripts/check_migration_safety.py

@@ -0,0 +1,209 @@
+#!/usr/bin/env python3
+"""
+数据库迁移破坏性变更安全检查
+
+只检查 Alembic 迁移文件的 upgrade 函数中是否包含破坏性操作：
+- DROP TABLE
+- ALTER TABLE ... DROP COLUMN
+- 列类型变更（可能导致数据丢失）
+- NOT NULL 约束新增（无默认值时）
+- RENAME TABLE / RENAME COLUMN
+
+忽略 downgrade 函数中的操作（那是回滚逻辑，正常的）。
+
+退出码：
+  0 - 安全 / 只有非破坏性变更
+  1 - 检测到高风险破坏性变更
+  2 - 检测到中风险变更，需人工确认
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import re
+import sys
+from pathlib import Path
+from typing import List, Tuple
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+ALEMBIC_VERSIONS_DIR = REPO_ROOT / "alembic" / "versions"
+
+# 高风险模式：直接导致数据丢失（只在 upgrade 中检查）
+HIGH_RISK_PATTERNS = [
+    (r"\bop\.drop_table\(", "op.drop_table() - 删除表，数据永久丢失"),
+    (r"\bop\.drop_column\(", "op.drop_column() - 删除列，数据永久丢失"),
+]
+
+# 中风险模式：可能导致数据丢失或兼容性问题
+MEDIUM_RISK_PATTERNS = [
+    (r"op\.alter_column\([^)]*nullable\s*=\s*False", "新增 NOT NULL 约束 - 旧数据可能为空导致迁移失败"),
+    (r"op\.alter_column\([^)]*type_\s*=", "列类型变更 - 可能导致数据截断或转换失败"),
+    (r"\bop\.rename_table\(", "op.rename_table() - 重命名表，可能导致依赖该表的代码报错"),
+    (r"\bop\.rename_column\(", "op.rename_column() - 重命名列，可能导致依赖该列的代码报错"),
+    (r"\bop\.drop_index\(", "op.drop_index() - 删除索引，可能影响查询性能"),
+    (r"\bop\.drop_constraint\(", "op.drop_constraint() - 删除约束，可能影响数据完整性"),
+]
+
+# 安全模式：这些是安全的新增操作
+SAFE_PATTERNS = [
+    (r"\bop\.create_table\(", "新建表"),
+    (r"\bop\.add_column\(", "新增列"),
+    (r"\bop\.create_index\(", "新建索引"),
+    (r"\bop\.create_unique_constraint\(", "新建唯一约束"),
+    (r"\bop\.create_foreign_key\(", "新建外键约束"),
+]
+
+
+def extract_upgrade_content(content: str) -> str:
+    """
+    从迁移文件中提取 upgrade 函数的内容。
+    只检查 upgrade 中的操作，忽略 downgrade。
+    """
+    # 匹配 def upgrade(): 或 def upgrade() -> None: 等格式
+    upgrade_match = re.search(r"def upgrade\b[^:]*:", content)
+    if not upgrade_match:
+        return ""
+
+    upgrade_start = upgrade_match.end()
+
+    # 找到下一个顶层 def（通常是 def downgrade）作为结束位置
+    rest = content[upgrade_start:]
+    downgrade_match = re.search(r"\n\ndef\s+\w+\b", rest)
+    if downgrade_match:
+        upgrade_end = upgrade_start + downgrade_match.start()
+    else:
+        upgrade_end = len(content)
+
+    return content[upgrade_start:upgrade_end]
+
+
+def find_new_migrations(since_revision: str | None = None) -> List[Path]:
+    """
+    找出新增的迁移文件。
+    如果指定了 since_revision，则找出该版本之后的所有迁移；
+    否则找出所有迁移文件。
+    """
+    all_migrations = sorted(ALEMBIC_VERSIONS_DIR.glob("*.py"))
+    if not since_revision:
+        return all_migrations
+
+    result = []
+    found = False
+    for m in all_migrations:
+        if since_revision in m.name or since_revision in m.stem:
+            found = True
+            continue
+        if found:
+            result.append(m)
+
+    return result if found else all_migrations
+
+
+def analyze_migration(file_path: Path) -> Tuple[List[str], List[str], List[str]]:
+    """分析单个迁移文件 upgrade 部分的风险等级"""
+    content = file_path.read_text()
+    upgrade_content = extract_upgrade_content(content)
+
+    if not upgrade_content:
+        return [], [], [f"{file_path.name}: 未找到 upgrade 函数"]
+
+    high_risks = []
+    medium_risks = []
+    safes = []
+
+    for pattern, desc in HIGH_RISK_PATTERNS:
+        if re.search(pattern, upgrade_content):
+            high_risks.append(f"{file_path.name}: {desc}")
+
+    for pattern, desc in MEDIUM_RISK_PATTERNS:
+        if re.search(pattern, upgrade_content):
+            medium_risks.append(f"{file_path.name}: {desc}")
+
+    for pattern, desc in SAFE_PATTERNS:
+        if re.search(pattern, upgrade_content):
+            safes.append(f"{file_path.name}: {desc}")
+
+    return high_risks, medium_risks, safes
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    parser.add_argument(
+        "--since",
+        default=os.getenv("MIGRATION_SINCE_REVISION"),
+        help="只检查指定版本之后的迁移（如：030_xxx），不传则检查所有迁移",
+    )
+    parser.add_argument(
+        "--warn-only",
+        action="store_true",
+        help="只警告不失败（用于非强制门禁场景）",
+    )
+    parser.add_argument(
+        "--allow-medium-risk",
+        action="store_true",
+        help="允许中风险变更（只拦截高风险）",
+    )
+    args = parser.parse_args()
+
+    migrations = find_new_migrations(args.since)
+
+    if not migrations:
+        print("✅ 未找到需要检查的迁移文件")
+        return 0
+
+    print(f"🔍 正在检查 {len(migrations)} 个迁移文件的 upgrade 操作...")
+    print()
+
+    all_high = []
+    all_medium = []
+    all_safe = []
+
+    for m in migrations:
+        high, medium, safe = analyze_migration(m)
+        all_high.extend(high)
+        all_medium.extend(medium)
+        all_safe.extend(safe)
+
+    if all_safe:
+        print("✅ 安全变更：")
+        for s in all_safe:
+            print(f"   - {s}")
+        print()
+
+    if all_medium:
+        print("⚠️  中风险变更（需人工确认）：")
+        for m_item in all_medium:
+            print(f"   - {m_item}")
+        print()
+
+    if all_high:
+        print("❌ 高风险破坏性变更（禁止自动部署）：")
+        for h in all_high:
+            print(f"   - {h}")
+        print()
+
+    print("=" * 60)
+    print(f"检查结果：{len(all_safe)} 项安全 / {len(all_medium)} 项中风险 / {len(all_high)} 项高风险")
+    print()
+
+    if all_high:
+        print("❌ 检测到高风险破坏性变更，CI 检查失败！")
+        print("   如果确认这是预期操作，请在 MR/PR 中说明原因并获得审批。")
+        if args.warn_only:
+            return 0
+        return 1
+
+    if all_medium and not args.allow_medium_risk:
+        print("⚠️  检测到中风险变更，请人工确认后再部署。")
+        if args.warn_only:
+            return 0
+        print("（如需仅拦截高风险，可使用 --allow-medium-risk 参数）")
+        return 2
+
+    print("✅ 未检测到破坏性变更")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())

scripts/rollback_production.sh

@@ -0,0 +1,288 @@
+#!/bin/bash
+set -eu
+
+# ============================================================
+# 生产环境一键回滚脚本
+#
+# 用法:
+#   IMAGE_TAG=<版本号> REGISTRY_TOKEN=<token> sh rollback_production.sh
+#
+# 功能:
+#   1. 拉取指定版本镜像
+#   2. 数据库回滚到对应版本（alembic downgrade）
+#   3. 重启 api/worker/web 三个服务
+#   4. 健康检查确认服务正常
+#
+# 环境变量:
+#   IMAGE_TAG           - 要回滚到的版本标签（必填）
+#   REGISTRY_TOKEN      - Registry 访问 token（可选）
+#   SKIP_DB_ROLLBACK    - 跳过数据库回滚（1=跳过，默认不跳过）
+#   DB_ROLLBACK_REV     - 数据库回滚到的版本（默认自动用镜像里的 head）
+# ============================================================
+
+IMAGE_TAG="${IMAGE_TAG:-}"
+REGISTRY="${REGISTRY:-git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas}"
+REGISTRY_USER="${REGISTRY_USER:-xiaoxia}"
+REGISTRY_TOKEN="${REGISTRY_TOKEN:-}"
+SKIP_DB_ROLLBACK="${SKIP_DB_ROLLBACK:-0}"
+DB_ROLLBACK_REV="${DB_ROLLBACK_REV:-}"
+
+ENV_FILE="${ENV_FILE:-/var/lib/xiaoxia-saas-production/.env}"
+GENERATED_DIR="${GENERATED_DIR:-/var/lib/xiaoxia-saas-production/generated}"
+LEGACY_ASSETS_DIR="${LEGACY_ASSETS_DIR:-/var/lib/xiaoxia-saas-production/legacy-assets}"
+
+if [ -z "$IMAGE_TAG" ]; then
+  echo "❌ IMAGE_TAG 是必填参数"
+  echo "用法: IMAGE_TAG=v0.1.125 REGISTRY_TOKEN=xxx sh rollback_production.sh"
+  exit 1
+fi
+
+test -f "$ENV_FILE"
+mkdir -p "$GENERATED_DIR"
+mkdir -p "$LEGACY_ASSETS_DIR"
+
+echo "=========================================="
+echo "  生产环境回滚  →  $IMAGE_TAG"
+echo "=========================================="
+echo ""
+
+# 先获取当前版本
+CURRENT_VERSION=""
+if docker inspect xiaoxia-api-production >/dev/null 2>&1; then
+  CURRENT_VERSION=$(docker inspect --format '{{ index .Config.Env 0 }}' xiaoxia-api-production 2>/dev/null | grep APP_VERSION | cut -d= -f2 || echo "unknown")
+fi
+echo "当前版本: ${CURRENT_VERSION:-unknown}"
+echo "回滚目标: $IMAGE_TAG"
+echo ""
+
+# 确认
+read -p "⚠️  确认要回滚生产环境到 $IMAGE_TAG 吗？(输入 YES 确认): " confirm
+if [ "$confirm" != "YES" ]; then
+  echo "已取消"
+  exit 0
+fi
+echo ""
+
+# ----- 登录 Registry -----
+if [ -n "$REGISTRY_TOKEN" ]; then
+  echo "登录 Registry: $REGISTRY"
+  REGISTRY_HOST=$(echo "$REGISTRY" | cut -d/ -f1)
+  printf '%s' "$REGISTRY_TOKEN" | docker login "$REGISTRY_HOST" -u "$REGISTRY_USER" --password-stdin 2>/dev/null || {
+    echo "WARN: docker login failed, will try to pull anyway"
+  }
+fi
+
+# ----- Pull 镜像 -----
+REGISTRY_API="${REGISTRY}/xiaoxia-saas-api:${IMAGE_TAG}"
+REGISTRY_WORKER="${REGISTRY}/xiaoxia-saas-worker:${IMAGE_TAG}"
+REGISTRY_WEB="${REGISTRY}/xiaoxia-saas-web:${IMAGE_TAG}"
+
+LOCAL_API="xiaoxia-saas-api:${IMAGE_TAG}"
+LOCAL_WORKER="xiaoxia-saas-worker:${IMAGE_TAG}"
+LOCAL_WEB="xiaoxia-saas-web:${IMAGE_TAG}"
+
+echo "Pulling API image..."
+docker pull "$REGISTRY_API"
+echo "Pulling Worker image..."
+docker pull "$REGISTRY_WORKER"
+echo "Pulling Web image..."
+docker pull "$REGISTRY_WEB"
+
+docker tag "$REGISTRY_API" "$LOCAL_API"
+docker tag "$REGISTRY_WORKER" "$LOCAL_WORKER"
+docker tag "$REGISTRY_WEB" "$LOCAL_WEB"
+echo "所有镜像拉取完成"
+echo ""
+
+# ----- 检查基础设施容器 -----
+echo "检查基础设施容器..."
+for c in xiaoxia-postgres-production xiaoxia-redis-production; do
+  if ! docker inspect "$c" >/dev/null 2>&1; then
+    echo "ERROR: Required container not found: $c"
+    exit 1
+  fi
+  state=$(docker inspect -f '{{.State.Status}}' "$c")
+  if [ "$state" != "running" ]; then
+    echo "ERROR: Container not running: $c ($state)"
+    exit 1
+  fi
+done
+
+docker network create xiaoxia-net-production 2>/dev/null || true
+echo ""
+
+# ----- 数据库回滚 -----
+if [ "$SKIP_DB_ROLLBACK" = "1" ]; then
+  echo "⏭️  跳过数据库回滚（SKIP_DB_ROLLBACK=1）"
+else
+  echo "🔄 执行数据库回滚..."
+  if [ -n "$DB_ROLLBACK_REV" ]; then
+    # 回滚到指定版本
+    echo "回滚到版本: $DB_ROLLBACK_REV"
+    docker run --rm \
+      --env-file "$ENV_FILE" \
+      --network xiaoxia-net-production \
+      -e APP_ENV=production \
+      "$LOCAL_API" sh -c "cd /app && alembic downgrade $DB_ROLLBACK_REV"
+  else
+    # 用目标镜像的 alembic head 来判断是否需要回滚
+    # 先检查当前DB版本和目标版本的关系
+    echo "检测数据库当前版本与目标版本..."
+    CURRENT_DB_REV=$(docker run --rm \
+      --env-file "$ENV_FILE" \
+      --network xiaoxia-net-production \
+      -e APP_ENV=production \
+      "$LOCAL_API" sh -c "cd /app && alembic current" 2>&1 | tail -1 | awk '{print $1}')
+    TARGET_DB_HEAD=$(docker run --rm \
+      "$LOCAL_API" sh -c "cd /app && alembic head" 2>&1 | tail -1 | awk '{print $1}')
+
+    echo "当前 DB 版本: ${CURRENT_DB_REV:-unknown}"
+    echo "目标 DB 版本: ${TARGET_DB_HEAD:-unknown}"
+
+    if [ "$CURRENT_DB_REV" = "$TARGET_DB_HEAD" ]; then
+      echo "✅ 数据库版本与目标版本一致，无需回滚"
+    else
+      echo "⚠️  数据库版本不一致，尝试回滚..."
+      echo "注意：自动回滚可能无法正确处理，请确认 DB_ROLLBACK_REV 参数"
+      echo "如果需要跳过数据库回滚，请设置 SKIP_DB_ROLLBACK=1"
+      exit 1
+    fi
+  fi
+  echo "数据库回滚完成"
+fi
+echo ""
+
+# ----- 停止旧容器 -----
+echo "停止旧容器..."
+docker rm -f xiaoxia-api-production 2>/dev/null || true
+docker rm -f xiaoxia-worker-production 2>/dev/null || true
+docker rm -f xiaoxia-web-production 2>/dev/null || true
+echo ""
+
+# ----- 启动新容器 -----
+LOG_OPTS="--log-driver json-file --log-opt max-size=50m --log-opt max-file=3"
+
+echo "启动 API 容器..."
+docker run -d \
+  --name xiaoxia-api-production \
+  --env-file "$ENV_FILE" \
+  --network xiaoxia-net-production \
+  -p 127.0.0.1:8001:8000 \
+  -e APP_ENV=production \
+  -e APP_VERSION="$IMAGE_TAG" \
+  -e GENERATED_FILES_DIR=/app/generated \
+  -e GENERATED_FILES_URL_PREFIX=/generated-files \
+  -e PUBLIC_API_BASE_URL=https://api.xiaoxiajiianji.com \
+  -v "$GENERATED_DIR:/app/generated" \
+  --restart unless-stopped \
+  --cpus 2 \
+  --memory 2g \
+  --health-cmd "python -c \"import urllib.request; urllib.request.urlopen('http://localhost:8000/health', timeout=5)\"" \
+  --health-interval 30s \
+  --health-timeout 10s \
+  --health-retries 3 \
+  --health-start-period 40s \
+  $LOG_OPTS \
+  "$LOCAL_API"
+
+echo "启动 Worker 容器..."
+docker run -d \
+  --name xiaoxia-worker-production \
+  --env-file "$ENV_FILE" \
+  --network xiaoxia-net-production \
+  -e APP_ENV=production \
+  -e APP_VERSION="$IMAGE_TAG" \
+  -e WORKER_CONCURRENCY=1 \
+  -e WORKER_MAX_TASKS_PER_CHILD=100 \
+  -e GENERATED_FILES_DIR=/app/generated \
+  -e GENERATED_FILES_URL_PREFIX=/generated-files \
+  -e PUBLIC_API_BASE_URL=https://api.xiaoxiajiianji.com \
+  -v "$GENERATED_DIR:/app/generated" \
+  --restart unless-stopped \
+  --cpus 2 \
+  --memory 2g \
+  --health-cmd "sh -c \"grep -q celery /proc/1/cmdline || exit 1\"" \
+  --health-interval 30s \
+  --health-timeout 10s \
+  --health-retries 3 \
+  --health-start-period 30s \
+  $LOG_OPTS \
+  "$LOCAL_WORKER"
+
+# Web legacy assets
+LEGACY_VOLUME=""
+if [ -d "$LEGACY_ASSETS_DIR" ] && [ "$(ls -A "$LEGACY_ASSETS_DIR" 2>/dev/null)" ]; then
+  LEGACY_VOLUME="-v ${LEGACY_ASSETS_DIR}:/usr/share/nginx/html/assets-legacy/assets:ro"
+  echo "Web 容器: legacy assets 已挂载"
+else
+  echo "Web 容器: 没有 legacy assets"
+fi
+
+echo "启动 Web 容器..."
+docker run -d \
+  --name xiaoxia-web-production \
+  --network xiaoxia-net-production \
+  -p 127.0.0.1:3002:80 \
+  --restart unless-stopped \
+  --cpus 0.5 \
+  --memory 512m \
+  $LEGACY_VOLUME \
+  --health-cmd "wget --spider -q http://127.0.0.1:80" \
+  --health-interval 30s \
+  --health-timeout 5s \
+  --health-retries 3 \
+  $LOG_OPTS \
+  "$LOCAL_WEB"
+echo ""
+
+# ----- 健康检查 -----
+echo "等待 API 健康..."
+i=0
+while [ "$i" -lt 40 ]; do
+  if curl -sf --max-time 5 http://127.0.0.1:8001/health >/dev/null 2>&1; then
+    echo "API is healthy!"
+    break
+  fi
+  i=$((i + 1))
+  echo "  Waiting... ($i/40)"
+  sleep 3
+done
+
+if [ "$i" -ge 40 ]; then
+  echo "❌ API 在 120s 内未就绪"
+  docker logs --tail 50 xiaoxia-api-production
+  exit 1
+fi
+
+echo "等待 Web 健康..."
+i=0
+while [ "$i" -lt 15 ]; do
+  if curl -sf --max-time 5 http://127.0.0.1:3002/ >/dev/null 2>&1; then
+    echo "Web is healthy!"
+    break
+  fi
+  i=$((i + 1))
+  echo "  Waiting... ($i/15)"
+  sleep 2
+done
+
+if [ "$i" -ge 15 ]; then
+  echo "❌ Web 在 30s 内未就绪"
+  docker logs --tail 30 xiaoxia-web-production
+  exit 1
+fi
+echo ""
+
+# ----- 清理 -----
+echo "清理旧镜像..."
+docker image prune -af --filter "until=168h" 2>/dev/null || true
+docker builder prune -af --filter "until=168h" 2>/dev/null || true
+
+echo ""
+echo "=========================================="
+echo "  ✅ 生产环境回滚完成"
+echo "=========================================="
+echo "API:      http://127.0.0.1:8001"
+echo "Web:      http://127.0.0.1:3002"
+echo "Version:  $IMAGE_TAG"
+docker ps --format "table {{.Names}}\t{{.Status}}\t{{.Image}}" | grep production