feat(ci): 新增Staging健康检查脚本

This commit is contained in:
2026-07-14 21:33:43 +08:00
committed by XiaoXia Bot
parent c540d88306
commit e45426fe7d
+328
View File
@@ -0,0 +1,328 @@
#!/bin/bash
# ===========================================
# CI Staging 健康检查 + 自动回滚脚本(Watchtower 模式)
# ===========================================
#
# 在 CI Runner 上执行,通过公网 URL 检查 Staging 部署健康状态。
# 不健康则自动回滚(把 :staging tag 指回旧版本镜像,Watchtower 检测到后自动回滚)。
#
# 用法:
# ./ci_staging_healthcheck.sh
#
# 环境变量:
# STAGING_API_URL - Staging API 地址 (默认 https://staging-api.xiaoxiajianji.com)
# STAGING_WEB_URL - Staging Web 地址 (默认 https://staging.xiaoxiajianji.com)
# HEALTH_CHECK_TIMEOUT - 健康检查总超时秒数 (默认 120)
# WATCHTOWER_WAIT - 等待 Watchtower 检测更新的秒数 (默认 90)
# SKIP_ROLLBACK - 失败时不自动回滚 (true/false, 默认 false)
# SKIP_NOTIFY - 跳过通知 (true/false, 默认 false)
# CI_NOTIFY_WEBHOOK - 通知 Webhook URL
#
# REGISTRY_TOKEN - Gitea Registry Token(回滚时需要)
# REGISTRY_HOST - Registry 地址 (默认 git.xiaoxiajianji.com)
# REGISTRY_REPO - 仓库路径 (默认 xiaoxia/xiaoxia-saas)
#
# GITHUB_SHA - 当前 commit SHA
# GITHUB_REF_NAME - 分支名
# GITHUB_RUN_ID - CI Run ID
# GITHUB_REPOSITORY - 仓库名
# GITHUB_ACTOR - 提交者
set -eu
SCRIPT_DIR="$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)"
# 配置
STAGING_API_URL="${STAGING_API_URL:-https://staging-api.xiaoxiajianji.com}"
STAGING_WEB_URL="${STAGING_WEB_URL:-https://staging.xiaoxiajianji.com}"
HEALTH_CHECK_TIMEOUT="${HEALTH_CHECK_TIMEOUT:-120}"
WATCHTOWER_WAIT="${WATCHTOWER_WAIT:-90}"
SKIP_ROLLBACK="${SKIP_ROLLBACK:-false}"
SKIP_NOTIFY="${SKIP_NOTIFY:-false}"
REGISTRY_HOST="${REGISTRY_HOST:-git.xiaoxiajianji.com}"
REGISTRY_REPO="${REGISTRY_REPO:-xiaoxia/xiaoxia-saas}"
# 颜色
RED='\033[0;31m'
GREEN='\033[0;32m'
YELLOW='\033[1;33m'
BLUE='\033[0;34m'
NC='\033[0m'
log_info() { echo -e "${GREEN}[INFO]${NC} $1"; }
log_warn() { echo -e "${YELLOW}[WARN]${NC} $1"; }
log_error() { echo -e "${RED}[ERROR]${NC} $1"; }
log_step() { echo -e "${BLUE}[STEP]${NC} $1"; }
# ===========================================
# 1. 记录部署前 :staging 镜像 digest(用于回滚)
# ===========================================
save_rollback_target() {
log_step "记录当前 :staging 镜像 digest(回滚目标)..."
# 登录 Registry
if [ -n "${REGISTRY_TOKEN:-}" ]; then
printf '%s' "${REGISTRY_TOKEN}" | docker login "${REGISTRY_HOST}" -u xiaoxia --password-stdin 2>/dev/null || true
fi
ROLLBACK_IMAGES=""
for svc in api worker web; do
image="${REGISTRY_HOST}/${REGISTRY_REPO}/xiaoxia-saas-${svc}:staging"
# pull 当前 :staging 镜像获取 digest
if docker pull "$image" 2>/dev/null; then
digest=$(docker inspect -f '{{index .RepoDigests 0}}' "$image" 2>/dev/null || echo "")
if [ -n "$digest" ]; then
ROLLBACK_IMAGES="${ROLLBACK_IMAGES}${svc}:${digest} "
log_info " $svc: $digest"
else
log_warn " $svc: 无法获取 digest,该服务将不参与回滚"
fi
else
log_warn " $svc: 拉取 :staging 镜像失败,该服务将不参与回滚"
fi
done
if [ -z "$ROLLBACK_IMAGES" ]; then
log_warn "没有获取到任何可回滚镜像,回滚功能将不可用"
fi
export ROLLBACK_IMAGES
}
# ===========================================
# 2. 等待 Watchtower 更新
# ===========================================
wait_for_watchtower() {
log_step "等待 Watchtower 检测新镜像并更新(${WATCHTOWER_WAIT}s..."
sleep "$WATCHTOWER_WAIT"
log_info "等待结束,开始健康检查"
}
# ===========================================
# 3. 健康检查
# ===========================================
health_check() {
local timeout="$HEALTH_CHECK_TIMEOUT"
local start_time
start_time=$(date +%s)
log_step "健康检查(超时 ${timeout}s..."
log_info " API: ${STAGING_API_URL}/health"
log_info " Web: ${STAGING_WEB_URL}/"
local api_ok=false
local web_ok=false
local api_docs_ok=false
local login_api_ok=false
while [ $(( $(date +%s) - start_time )) -lt "$timeout" ]; do
# 检查 API health
if [ "$api_ok" = false ] && curl -sf --max-time 10 "${STAGING_API_URL}/health" >/dev/null 2>&1; then
log_info "✅ API 健康检查通过"
api_ok=true
fi
# 检查 Web 首页
if [ "$web_ok" = false ] && curl -sf --max-time 10 "$STAGING_WEB_URL/" >/dev/null 2>&1; then
log_info "✅ Web 前端检查通过"
web_ok=true
fi
# 检查 API docs(服务完全启动的标志)
if [ "$api_docs_ok" = false ]; then
HTTP_CODE=$(curl -s -o /dev/null -w "%{http_code}" --max-time 10 "${STAGING_API_URL}/docs" 2>/dev/null || echo "000")
if [ "$HTTP_CODE" = "200" ]; then
log_info "✅ API Docs 检查通过"
api_docs_ok=true
fi
fi
# 检查登录 API(业务逻辑正常的标志)
if [ "$login_api_ok" = false ]; then
HTTP_CODE=$(curl -s -o /dev/null -w "%{http_code}" --max-time 10 -X POST \
"${STAGING_API_URL}/api/v1/auth/login" \
-H "Content-Type: application/json" \
-d '{"email":"smoke@test.com","password":"wrong"}' 2>/dev/null || echo "000")
if [ "$HTTP_CODE" = "401" ] || [ "$HTTP_CODE" = "422" ]; then
log_info "✅ 登录 API 检查通过(HTTP $HTTP_CODE,符合预期)"
login_api_ok=true
fi
fi
# 都通过了就退出
if [ "$api_ok" = true ] && [ "$web_ok" = true ] && [ "$api_docs_ok" = true ] && [ "$login_api_ok" = true ]; then
log_info "🎉 所有健康检查通过!"
return 0
fi
sleep 5
done
# 超时了
log_error "❌ 健康检查超时 (${timeout}s)"
[ "$api_ok" = false ] && log_error " - API health 未通过"
[ "$web_ok" = false ] && log_error " - Web 前端未通过"
[ "$api_docs_ok" = false ] && log_error " - API Docs 未通过"
[ "$login_api_ok" = false ] && log_error " - 登录 API 未通过"
return 1
}
# ===========================================
# 4. 执行回滚(把 :staging tag 改回旧版本)
# ===========================================
do_rollback() {
log_step "执行回滚:将 :staging tag 指回旧版本镜像..."
if [ -z "${ROLLBACK_IMAGES:-}" ]; then
log_error "没有可回滚的镜像记录,无法自动回滚"
return 1
fi
local rollback_ok=true
for entry in $ROLLBACK_IMAGES; do
svc="${entry%%:*}"
digest="${entry#*:}"
# digest 是 host/repo/image@sha256:xxx 格式,entry 里只取了第一部分,需要重新组合
# 实际上 ROLLBACK_IMAGES 存的是 svc:repo/image@sha256:xxx
# 让我们重新从完整digest中提取
image_digest="${entry#*:}:${entry#*:*:}"
# 不对,格式是 svc:REGISTRY_HOST/REGISTRY_REPO/xiaoxia-saas-svc@sha256:xxx
# 让我重新组织一下
:
done
# 上面的解析有问题,重新来
# ROLLBACK_IMAGES 格式: "api:digest1 worker:digest2 web:digest3"
# 但digest本身带冒号(@sha256:xxx),所以分隔有问题
# 改用数组方式
log_info "开始回滚各服务..."
local IFS=' '
for pair in $ROLLBACK_IMAGES; do
svc="${pair%%:*}"
# 剩余部分是完整的 digest 字符串(含 @sha256:xxx
digest="${pair#*:}"
image_name="${REGISTRY_HOST}/${REGISTRY_REPO}/xiaoxia-saas-${svc}"
log_info " 回滚 $svc 到: $digest"
# 通过 digest 拉取旧镜像
if docker pull "$digest" 2>/dev/null; then
# 重新打 :staging tag
docker tag "$digest" "${image_name}:staging"
docker push "${image_name}:staging"
log_info "$svc 已回滚"
else
log_error "$svc 回滚失败:无法拉取旧镜像"
rollback_ok=false
fi
done
if [ "$rollback_ok" = true ]; then
log_info "所有服务回滚命令执行完成,Watchtower 将在 ~60s 内检测到并更新"
return 0
else
log_error "部分服务回滚失败"
return 1
fi
}
# ===========================================
# 5. 发送通知
# ===========================================
send_notification() {
local status="$1" # success / failure / rollback
local detail="$2"
if [ "${SKIP_NOTIFY:-false}" = "true" ]; then
log_info "跳过通知(SKIP_NOTIFY=true"
return 0
fi
local webhook="${CI_NOTIFY_WEBHOOK:-}"
if [ -z "$webhook" ]; then
log_warn "未配置 CI_NOTIFY_WEBHOOK,跳过通知"
return 0
fi
# 调用通知脚本
if [ -f "$SCRIPT_DIR/deploy_notify.py" ]; then
python3 "$SCRIPT_DIR/deploy_notify.py" \
--status "$status" \
--detail "$detail" \
--webhook "$webhook" \
2>/dev/null || log_warn "通知发送失败(非致命)"
else
log_warn "找不到 deploy_notify.py,跳过通知"
fi
}
# ===========================================
# 主流程
# ===========================================
main() {
echo ""
echo "=========================================="
echo " CI Staging 健康检查 + 自动回滚"
echo "=========================================="
echo ""
local deploy_status="success"
local deploy_detail=""
# 1. 记录部署前状态(回滚目标)
save_rollback_target
# 2. 等待 Watchtower 更新
wait_for_watchtower
# 3. 健康检查
if ! health_check; then
log_error "健康检查失败"
deploy_status="failure"
deploy_detail="健康检查超时,部署后服务未正常启动"
# 自动回滚
if [ "${SKIP_ROLLBACK:-false}" != "true" ]; then
log_warn "开始自动回滚..."
if do_rollback; then
deploy_status="rollback"
deploy_detail="健康检查失败,已自动回滚到上一版本"
# 回滚后再检查一下回滚后的状态
log_info "等待 Watchtower 应用回滚(90s..."
sleep 90
if health_check; then
log_info "✅ 回滚后服务已恢复"
deploy_detail="${deploy_detail},回滚后服务已恢复"
else
log_error "⚠️ 回滚后健康检查仍未通过,请手动排查"
deploy_detail="${deploy_detail},但回滚后仍未恢复,请紧急排查"
fi
else
deploy_detail="健康检查失败且回滚失败,请手动排查"
fi
fi
send_notification "$deploy_status" "$deploy_detail"
# 失败时退出非零,让 CI Job 标记为失败
exit 1
fi
# 4. 成功
log_info ""
log_info "=================================="
log_info " ✅ Staging 部署成功!"
log_info "=================================="
deploy_detail="部署成功,所有健康检查通过"
send_notification "success" "$deploy_detail"
}
main "$@"