feat(prometheus): 实现Alertmanager兼容API并重构告警处理

- 新增Alertmanager API v2兼容接口用于接收Prometheus告警
- 重构告警服务架构，替换原有的轮询模式为推送模式
- 添加docker-compose配置支持Prometheus管理API
- 移除过时的AlertWebhookService实现

Author: JaD1ng

internal/prometheus_adapter/api/alertmanager_api.go

@@ -0,0 +1,25 @@
+package api
+
+import (
+	"github.com/fox-gonic/fox"
+	"github.com/qiniu/zeroops/internal/prometheus_adapter/service"
+)
+
+// setupAlertmanagerRouters 设置 Alertmanager 兼容路由
+// 这些路由模拟 Alertmanager API，接收 Prometheus 的告警推送
+func (api *Api) setupAlertmanagerRouters(router *fox.Engine, alertmanagerService *service.AlertmanagerService) {
+	// Alertmanager API v2 告警接收端点
+	router.POST("/api/v2/alerts", func(c *fox.Context) {
+		alertmanagerService.HandleAlertsV2(c.Writer, c.Request)
+	})
+
+	// 健康检查端点
+	router.GET("/-/healthy", func(c *fox.Context) {
+		alertmanagerService.HandleHealthCheck(c.Writer, c.Request)
+	})
+
+	// 就绪检查端点
+	router.GET("/-/ready", func(c *fox.Context) {
+		alertmanagerService.HandleReadyCheck(c.Writer, c.Request)
+	})
+}

internal/prometheus_adapter/api/api.go

@@ -12,17 +12,19 @@ import (
 
 // Api Prometheus Adapter API
 type Api struct {
-	metricService *service.MetricService
-	alertService  *service.AlertService
-	router        *fox.Engine
+	metricService       *service.MetricService
+	alertService        *service.AlertService
+	alertmanagerService *service.AlertmanagerService
+	router              *fox.Engine
 }
 
 // NewApi 创建新的 API
-func NewApi(metricService *service.MetricService, alertService *service.AlertService, router *fox.Engine) (*Api, error) {
+func NewApi(metricService *service.MetricService, alertService *service.AlertService, alertmanagerService *service.AlertmanagerService, router *fox.Engine) (*Api, error) {
 	api := &Api{
-		metricService: metricService,
-		alertService:  alertService,
-		router:        router,
+		metricService:       metricService,
+		alertService:        alertService,
+		alertmanagerService: alertmanagerService,
+		router:              router,
 	}
 
 	api.setupRouters(router)
@@ -35,6 +37,8 @@ func (api *Api) setupRouters(router *fox.Engine) {
 	api.setupMetricRouters(router)
 	// 告警相关路由
 	api.setupAlertRouters(router)
+	// Alertmanager 兼容路由
+	api.setupAlertmanagerRouters(router, api.alertmanagerService)
 }
 
 // ========== 通用辅助方法 ==========

internal/prometheus_adapter/model/alert.go

@@ -17,3 +17,12 @@ type AlertRuleMeta struct {
 	Labels    string  `json:"labels" gorm:"type:jsonb"`                  // 适用标签，如 {"service":"s3","version":"v1"}，为空表示全局
 	Threshold float64 `json:"threshold"`                                 // 阈值（会被渲染成特定规则的 threshold metric 数值）
 }
+
+// AlertmanagerAlert 符合 Alertmanager API v2 的告警格式
+type AlertmanagerAlert struct {
+	Labels       map[string]string `json:"labels"`
+	Annotations  map[string]string `json:"annotations,omitempty"`
+	StartsAt     string            `json:"startsAt,omitempty"` // RFC3339 格式
+	EndsAt       string            `json:"endsAt,omitempty"`   // RFC3339 格式
+	GeneratorURL string            `json:"generatorURL,omitempty"`
+}

internal/prometheus_adapter/server.go

@@ -15,13 +15,13 @@ import (
 
 // PrometheusAdapterServer Prometheus Adapter 服务器
 type PrometheusAdapterServer struct {
-	config              *config.Config
-	promConfig          *promconfig.PrometheusAdapterConfig
-	promClient          *client.PrometheusClient
-	metricService       *service.MetricService
-	alertService        *service.AlertService
-	alertWebhookService *service.AlertWebhookService
-	api                 *api.Api
+	config                   *config.Config
+	promConfig               *promconfig.PrometheusAdapterConfig
+	promClient               *client.PrometheusClient
+	metricService            *service.MetricService
+	alertService             *service.AlertService
+	alertmanagerProxyService *service.AlertmanagerService
+	api                      *api.Api
 }
 
 // NewPrometheusAdapterServer 创建新的 Prometheus Adapter 服务器
@@ -44,22 +44,16 @@ func NewPrometheusAdapterServer(cfg *config.Config) (*PrometheusAdapterServer, e
 	// 创建告警服务
 	alertService := service.NewAlertService(promClient, promConfig)
 
-	// 创建告警 Webhook 服务
-	alertWebhookService := service.NewAlertWebhookService(promClient, promConfig)
+	// 创建 Alertmanager 代理服务
+	alertmanagerProxyService := service.NewAlertmanagerProxyService(promConfig)
 
 	server := &PrometheusAdapterServer{
-		config:              cfg,
-		promConfig:          promConfig,
-		promClient:          promClient,
-		metricService:       metricService,
-		alertService:        alertService,
-		alertWebhookService: alertWebhookService,
-	}
-
-	// 启动告警 Webhook 服务
-	if err := alertWebhookService.Start(); err != nil {
-		log.Error().Err(err).Msg("Failed to start alert webhook service")
-		// 不返回错误，允许服务继续运行
+		config:                   cfg,
+		promConfig:               promConfig,
+		promClient:               promClient,
+		metricService:            metricService,
+		alertService:             alertService,
+		alertmanagerProxyService: alertmanagerProxyService,
 	}
 
 	log.Info().Str("prometheus_address", promConfig.Prometheus.Address).Msg("Prometheus Adapter initialized successfully")
@@ -69,23 +63,20 @@ func NewPrometheusAdapterServer(cfg *config.Config) (*PrometheusAdapterServer, e
 // UseApi 设置 API 路由
 func (s *PrometheusAdapterServer) UseApi(router *fox.Engine) error {
 	var err error
-	s.api, err = api.NewApi(s.metricService, s.alertService, router)
+	s.api, err = api.NewApi(s.metricService, s.alertService, s.alertmanagerProxyService, router)
 	if err != nil {
 		return fmt.Errorf("failed to initialize API: %w", err)
 	}
 
+	log.Info().Msg("All API endpoints registered")
+
 	return nil
 }
 
 // Close 优雅关闭服务器
 func (s *PrometheusAdapterServer) Close(ctx context.Context) error {
 	log.Info().Msg("Starting shutdown...")
 
-	// 停止告警 Webhook 服务
-	if s.alertWebhookService != nil {
-		s.alertWebhookService.Stop()
-	}
-
 	// 调用 alertService 的 Shutdown 方法保存规则
 	if s.alertService != nil {
 		if err := s.alertService.Shutdown(); err != nil {

internal/prometheus_adapter/service/alert_service.go

@@ -411,7 +411,7 @@ func (s *AlertService) buildExpression(rule *model.AlertRule, meta *model.AlertR
 		if len(labelMatchers) > 0 {
 			// 如果表达式包含{，说明已经有标签选择器
 			if strings.Contains(expr, "{") {
-				expr = strings.Replace(expr, "}", ","+strings.Join(labelMatchers, ",")+"}}", 1)
+				expr = strings.Replace(expr, "}", ","+strings.Join(labelMatchers, ",")+"}", 1)
 			} else {
 				// 在指标名后添加标签选择器
 				// 查找第一个非字母数字下划线的字符