feat(Metrics&Alert): 补充写入service_states缓存

Author: acd19ml

internal/alerting/service/receiver/README.md

@@ -170,9 +170,12 @@ func (h *Handler) AlertmanagerWebhook(c *gin.Context) {
             // 仅记录错误，不阻断主流程
         }
         // 6) 写通到 Redis（不阻塞主流程，失败仅记录日志）
+        //    alert_issues
         if err := h.cache.WriteIssue(c, row, a); err != nil {
             // 仅记录错误，避免影响 Alertmanager 重试逻辑
         }
+        //    service_states
+        _ = h.cache.WriteServiceState(c, a.Labels["service"], a.Labels["service_version"], row.AlertSince, "Error")
         MarkSeen(key) // 记忆幂等键
         created++
     }
@@ -383,6 +386,10 @@ key 设计与 TTL：
 - alert:idemp:{fingerprint}|{startsAtRFC3339Nano} → "1"，TTL 10m（用于分布式幂等 SETNX）
 - alert:index:open → Set(issues...)，无 TTL（恢复时再移除）
 - alert:index:svc:{service}:open → Set(issues...)，无 TTL
+// service_states 缓存
+- service_state:{service}:{version} → JSON（service/version/report_at/health_state），TTL 3d
+- service_state:index:service:{service} → Set(keys)
+- service_state:index:health:{health_state} → Set(keys)
 
 cache.go（示例）：
 
@@ -446,6 +453,9 @@ redis-cli --raw keys 'alert:*'
 redis-cli --raw get alert:issue:<id>
 redis-cli --raw smembers alert:index:open | head -n 10
 redis-cli ttl alert:issue:<id>
+redis-cli --raw keys 'service_state:*'
+redis-cli --raw get service_state:serviceA:v1.3.7
+redis-cli --raw smembers service_state:index:health:Error
 ```
 
 ⸻

internal/alerting/service/receiver/cache.go

@@ -15,13 +15,17 @@ import (
 type AlertIssueCache interface {
 	WriteIssue(ctx context.Context, r *AlertIssueRow, a AMAlert) error
 	TryMarkIdempotent(ctx context.Context, a AMAlert) (bool, error)
+	WriteServiceState(ctx context.Context, service, version string, reportAt time.Time, healthState string) error
 }
 
 // NoopCache is a no-op implementation of AlertIssueCache.
 type NoopCache struct{}
 
 func (NoopCache) WriteIssue(ctx context.Context, r *AlertIssueRow, a AMAlert) error { return nil }
 func (NoopCache) TryMarkIdempotent(ctx context.Context, a AMAlert) (bool, error)    { return true, nil }
+func (NoopCache) WriteServiceState(ctx context.Context, service, version string, reportAt time.Time, healthState string) error {
+	return nil
+}
 
 // Cache implements AlertIssueCache using Redis.
 type Cache struct{ R *redis.Client }
@@ -79,3 +83,30 @@ func (c *Cache) TryMarkIdempotent(ctx context.Context, a AMAlert) (bool, error)
 	ok, err := c.R.SetNX(ctx, k, "1", 10*time.Minute).Result()
 	return ok, err
 }
+
+// WriteServiceState writes the service state snapshot into Redis and maintains simple indices.
+func (c *Cache) WriteServiceState(ctx context.Context, service, version string, reportAt time.Time, healthState string) error {
+	if c == nil || c.R == nil {
+		return nil
+	}
+	s := strings.TrimSpace(service)
+	v := strings.TrimSpace(version)
+	key := "service_state:" + s + ":" + v
+	payload := map[string]any{
+		"service":      s,
+		"version":      v,
+		"report_at":    reportAt,
+		"health_state": healthState,
+	}
+	b, _ := json.Marshal(payload)
+	pipe := c.R.Pipeline()
+	pipe.Set(ctx, key, b, 72*time.Hour)
+	if s != "" {
+		pipe.SAdd(ctx, "service_state:index:service:"+s, key)
+	}
+	if healthState != "" {
+		pipe.SAdd(ctx, "service_state:index:health:"+healthState, key)
+	}
+	_, err := pipe.Exec(ctx)
+	return err
+}

internal/alerting/service/receiver/handler.go

@@ -66,6 +66,7 @@ func (h *Handler) AlertmanagerWebhook(c *fox.Context) {
 			version := strings.TrimSpace(a.Labels["service_version"]) // optional
 			if service != "" {
 				_ = w.UpsertServiceState(c.Request.Context(), service, version, row.AlertSince, "Error")
+				_ = h.cache.WriteServiceState(c.Request.Context(), service, version, row.AlertSince, "Error")
 			}
 		}
 		// Write-through to cache. Errors are ignored to avoid impacting webhook ack.