Merge pull request #285 from IanShaw027/fix/ops-bug

feat(ops): 增强错误日志管理、告警静默和前端 UI 优化
2026-01-15 11:26:16 +08:00
parent 28de614dfb 5354ba3662
commit 27214f8657
42 changed files with 4039 additions and 1065 deletions
--- a/backend/internal/service/admin_service_bulk_update_test.go
+++ b/backend/internal/service/admin_service_bulk_update_test.go
@@ -12,9 +12,9 @@ import (

 type accountRepoStubForBulkUpdate struct {
 	accountRepoStub
-	bulkUpdateErr     error
-	bulkUpdateIDs     []int64
-	bindGroupErrByID  map[int64]error
+	bulkUpdateErr    error
+	bulkUpdateIDs    []int64
+	bindGroupErrByID map[int64]error
 }

 func (s *accountRepoStubForBulkUpdate) BulkUpdate(_ context.Context, ids []int64, _ AccountBulkUpdate) (int64, error) {
--- a/backend/internal/service/antigravity_gateway_service.go
+++ b/backend/internal/service/antigravity_gateway_service.go
@@ -564,6 +564,10 @@ urlFallbackLoop:
 			}

 			upstreamReq, err := antigravity.NewAPIRequestWithURL(ctx, baseURL, action, accessToken, geminiBody)
+			// Capture upstream request body for ops retry of this attempt.
+			if c != nil {
+				c.Set(OpsUpstreamRequestBodyKey, string(geminiBody))
+			}
 			if err != nil {
 				return nil, err
 			}
@@ -574,6 +578,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: 0,
 					Kind:               "request_error",
 					Message:            safeErr,
@@ -615,6 +620,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "retry",
@@ -645,6 +651,7 @@ urlFallbackLoop:
 					appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 						Platform:           account.Platform,
 						AccountID:          account.ID,
+						AccountName:        account.Name,
 						UpstreamStatusCode: resp.StatusCode,
 						UpstreamRequestID:  resp.Header.Get("x-request-id"),
 						Kind:               "retry",
@@ -697,6 +704,7 @@ urlFallbackLoop:
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  resp.Header.Get("x-request-id"),
 				Kind:               "signature_error",
@@ -740,6 +748,7 @@ urlFallbackLoop:
 					appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 						Platform:           account.Platform,
 						AccountID:          account.ID,
+						AccountName:        account.Name,
 						UpstreamStatusCode: 0,
 						Kind:               "signature_retry_request_error",
 						Message:            sanitizeUpstreamErrorMessage(retryErr.Error()),
@@ -770,6 +779,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: retryResp.StatusCode,
 					UpstreamRequestID:  retryResp.Header.Get("x-request-id"),
 					Kind:               kind,
@@ -817,6 +827,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "failover",
@@ -1371,6 +1382,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: 0,
 					Kind:               "request_error",
 					Message:            safeErr,
@@ -1412,6 +1424,7 @@ urlFallbackLoop:
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "retry",
@@ -1442,6 +1455,7 @@ urlFallbackLoop:
 					appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 						Platform:           account.Platform,
 						AccountID:          account.ID,
+						AccountName:        account.Name,
 						UpstreamStatusCode: resp.StatusCode,
 						UpstreamRequestID:  resp.Header.Get("x-request-id"),
 						Kind:               "retry",
@@ -1543,6 +1557,7 @@ urlFallbackLoop:
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  requestID,
 				Kind:               "failover",
@@ -1559,6 +1574,7 @@ urlFallbackLoop:
 		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 			Platform:           account.Platform,
 			AccountID:          account.ID,
+			AccountName:        account.Name,
 			UpstreamStatusCode: resp.StatusCode,
 			UpstreamRequestID:  requestID,
 			Kind:               "http_error",
@@ -2039,6 +2055,7 @@ func (s *AntigravityGatewayService) writeMappedClaudeError(c *gin.Context, accou
 	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 		Platform:           account.Platform,
 		AccountID:          account.ID,
+		AccountName:        account.Name,
 		UpstreamStatusCode: upstreamStatus,
 		UpstreamRequestID:  upstreamRequestID,
 		Kind:               "http_error",
--- a/backend/internal/service/gateway_service.go
+++ b/backend/internal/service/gateway_service.go
@@ -1466,6 +1466,9 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 	for attempt := 1; attempt <= maxRetryAttempts; attempt++ {
 		// 构建上游请求（每次重试需要重新构建，因为请求体需要重新读取）
 		upstreamReq, err := s.buildUpstreamRequest(ctx, c, account, body, token, tokenType, reqModel)
+		// Capture upstream request body for ops retry of this attempt.
+		c.Set(OpsUpstreamRequestBodyKey, string(body))
+
 		if err != nil {
 			return nil, err
 		}
@@ -1482,6 +1485,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: 0,
 				Kind:               "request_error",
 				Message:            safeErr,
@@ -1506,6 +1510,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 					appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 						Platform:           account.Platform,
 						AccountID:          account.ID,
+						AccountName:        account.Name,
 						UpstreamStatusCode: resp.StatusCode,
 						UpstreamRequestID:  resp.Header.Get("x-request-id"),
 						Kind:               "signature_error",
@@ -1557,6 +1562,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 								appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 									Platform:           account.Platform,
 									AccountID:          account.ID,
+									AccountName:        account.Name,
 									UpstreamStatusCode: retryResp.StatusCode,
 									UpstreamRequestID:  retryResp.Header.Get("x-request-id"),
 									Kind:               "signature_retry_thinking",
@@ -1585,6 +1591,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 										appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 											Platform:           account.Platform,
 											AccountID:          account.ID,
+											AccountName:        account.Name,
 											UpstreamStatusCode: 0,
 											Kind:               "signature_retry_tools_request_error",
 											Message:            sanitizeUpstreamErrorMessage(retryErr2.Error()),
@@ -1643,6 +1650,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "retry",
@@ -1691,6 +1699,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  resp.Header.Get("x-request-id"),
 				Kind:               "retry_exhausted_failover",
@@ -1757,6 +1766,7 @@ func (s *GatewayService) Forward(ctx context.Context, c *gin.Context, account *A
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  resp.Header.Get("x-request-id"),
 					Kind:               "failover_on_400",
--- a/backend/internal/service/gemini_messages_compat_service.go
+++ b/backend/internal/service/gemini_messages_compat_service.go
@@ -545,12 +545,19 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 		}
 		requestIDHeader = idHeader

+		// Capture upstream request body for ops retry of this attempt.
+		if c != nil {
+			// In this code path `body` is already the JSON sent to upstream.
+			c.Set(OpsUpstreamRequestBodyKey, string(body))
+		}
+
 		resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
 		if err != nil {
 			safeErr := sanitizeUpstreamErrorMessage(err.Error())
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: 0,
 				Kind:               "request_error",
 				Message:            safeErr,
@@ -588,6 +595,7 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  upstreamReqID,
 					Kind:               "signature_error",
@@ -662,6 +670,7 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  upstreamReqID,
 					Kind:               "retry",
@@ -711,6 +720,7 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  upstreamReqID,
 				Kind:               "failover",
@@ -737,6 +747,7 @@ func (s *GeminiMessagesCompatService) Forward(ctx context.Context, c *gin.Contex
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  upstreamReqID,
 				Kind:               "failover",
@@ -972,12 +983,19 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 		}
 		requestIDHeader = idHeader

+		// Capture upstream request body for ops retry of this attempt.
+		if c != nil {
+			// In this code path `body` is already the JSON sent to upstream.
+			c.Set(OpsUpstreamRequestBodyKey, string(body))
+		}
+
 		resp, err = s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
 		if err != nil {
 			safeErr := sanitizeUpstreamErrorMessage(err.Error())
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: 0,
 				Kind:               "request_error",
 				Message:            safeErr,
@@ -1036,6 +1054,7 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 				appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 					Platform:           account.Platform,
 					AccountID:          account.ID,
+					AccountName:        account.Name,
 					UpstreamStatusCode: resp.StatusCode,
 					UpstreamRequestID:  upstreamReqID,
 					Kind:               "retry",
@@ -1120,6 +1139,7 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  requestID,
 				Kind:               "failover",
@@ -1143,6 +1163,7 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  requestID,
 				Kind:               "failover",
@@ -1168,6 +1189,7 @@ func (s *GeminiMessagesCompatService) ForwardNative(ctx context.Context, c *gin.
 		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 			Platform:           account.Platform,
 			AccountID:          account.ID,
+			AccountName:        account.Name,
 			UpstreamStatusCode: resp.StatusCode,
 			UpstreamRequestID:  requestID,
 			Kind:               "http_error",
@@ -1300,6 +1322,7 @@ func (s *GeminiMessagesCompatService) writeGeminiMappedError(c *gin.Context, acc
 	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 		Platform:           account.Platform,
 		AccountID:          account.ID,
+		AccountName:        account.Name,
 		UpstreamStatusCode: upstreamStatus,
 		UpstreamRequestID:  upstreamRequestID,
 		Kind:               "http_error",
--- a/backend/internal/service/openai_gateway_service.go
+++ b/backend/internal/service/openai_gateway_service.go
@@ -664,6 +664,11 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		proxyURL = account.Proxy.URL()
 	}

+	// Capture upstream request body for ops retry of this attempt.
+	if c != nil {
+		c.Set(OpsUpstreamRequestBodyKey, string(body))
+	}
+
 	// Send request
 	resp, err := s.httpUpstream.Do(upstreamReq, proxyURL, account.ID, account.Concurrency)
 	if err != nil {
@@ -673,6 +678,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 			Platform:           account.Platform,
 			AccountID:          account.ID,
+			AccountName:        account.Name,
 			UpstreamStatusCode: 0,
 			Kind:               "request_error",
 			Message:            safeErr,
@@ -707,6 +713,7 @@ func (s *OpenAIGatewayService) Forward(ctx context.Context, c *gin.Context, acco
 			appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 				Platform:           account.Platform,
 				AccountID:          account.ID,
+				AccountName:        account.Name,
 				UpstreamStatusCode: resp.StatusCode,
 				UpstreamRequestID:  resp.Header.Get("x-request-id"),
 				Kind:               "failover",
@@ -864,6 +871,7 @@ func (s *OpenAIGatewayService) handleErrorResponse(ctx context.Context, resp *ht
 		appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 			Platform:           account.Platform,
 			AccountID:          account.ID,
+			AccountName:        account.Name,
 			UpstreamStatusCode: resp.StatusCode,
 			UpstreamRequestID:  resp.Header.Get("x-request-id"),
 			Kind:               "http_error",
@@ -894,6 +902,7 @@ func (s *OpenAIGatewayService) handleErrorResponse(ctx context.Context, resp *ht
 	appendOpsUpstreamError(c, OpsUpstreamErrorEvent{
 		Platform:           account.Platform,
 		AccountID:          account.ID,
+		AccountName:        account.Name,
 		UpstreamStatusCode: resp.StatusCode,
 		UpstreamRequestID:  resp.Header.Get("x-request-id"),
 		Kind:               kind,
--- a/backend/internal/service/ops_alert_evaluator_service.go
+++ b/backend/internal/service/ops_alert_evaluator_service.go
@@ -206,7 +206,7 @@ func (s *OpsAlertEvaluatorService) evaluateOnce(interval time.Duration) {
 			continue
 		}

-		scopePlatform, scopeGroupID := parseOpsAlertRuleScope(rule.Filters)
+		scopePlatform, scopeGroupID, scopeRegion := parseOpsAlertRuleScope(rule.Filters)

 		windowMinutes := rule.WindowMinutes
 		if windowMinutes <= 0 {
@@ -236,6 +236,17 @@ func (s *OpsAlertEvaluatorService) evaluateOnce(interval time.Duration) {
 				continue
 			}

+			// Scoped silencing: if a matching silence exists, skip creating a firing event.
+			if s.opsService != nil {
+				platform := strings.TrimSpace(scopePlatform)
+				region := scopeRegion
+				if platform != "" {
+					if ok, err := s.opsService.IsAlertSilenced(ctx, rule.ID, platform, scopeGroupID, region, now); err == nil && ok {
+						continue
+					}
+				}
+			}
+
 			latestEvent, err := s.opsRepo.GetLatestAlertEvent(ctx, rule.ID)
 			if err != nil {
 				log.Printf("[OpsAlertEvaluator] get latest event failed (rule=%d): %v", rule.ID, err)
@@ -359,9 +370,9 @@ func requiredSustainedBreaches(sustainedMinutes int, interval time.Duration) int
 	return required
 }

-func parseOpsAlertRuleScope(filters map[string]any) (platform string, groupID *int64) {
+func parseOpsAlertRuleScope(filters map[string]any) (platform string, groupID *int64, region *string) {
 	if filters == nil {
-		return "", nil
+		return "", nil, nil
 	}
 	if v, ok := filters["platform"]; ok {
 		if s, ok := v.(string); ok {
@@ -392,7 +403,15 @@ func parseOpsAlertRuleScope(filters map[string]any) (platform string, groupID *i
 			}
 		}
 	}
-	return platform, groupID
+	if v, ok := filters["region"]; ok {
+		if s, ok := v.(string); ok {
+			vv := strings.TrimSpace(s)
+			if vv != "" {
+				region = &vv
+			}
+		}
+	}
+	return platform, groupID, region
 }

 func (s *OpsAlertEvaluatorService) computeRuleMetric(
@@ -504,16 +523,6 @@ func (s *OpsAlertEvaluatorService) computeRuleMetric(
 			return 0, false
 		}
 		return overview.UpstreamErrorRate * 100, true
-	case "p95_latency_ms":
-		if overview.Duration.P95 == nil {
-			return 0, false
-		}
-		return float64(*overview.Duration.P95), true
-	case "p99_latency_ms":
-		if overview.Duration.P99 == nil {
-			return 0, false
-		}
-		return float64(*overview.Duration.P99), true
 	default:
 		return 0, false
 	}
--- a/backend/internal/service/ops_alert_models.go
+++ b/backend/internal/service/ops_alert_models.go
@@ -8,8 +8,9 @@ import "time"
 // with the existing ops dashboard frontend (backup style).

 const (
-	OpsAlertStatusFiring   = "firing"
-	OpsAlertStatusResolved = "resolved"
+	OpsAlertStatusFiring         = "firing"
+	OpsAlertStatusResolved       = "resolved"
+	OpsAlertStatusManualResolved = "manual_resolved"
 )

 type OpsAlertRule struct {
@@ -58,12 +59,32 @@ type OpsAlertEvent struct {
 	CreatedAt time.Time `json:"created_at"`
 }

+type OpsAlertSilence struct {
+	ID int64 `json:"id"`
+
+	RuleID   int64   `json:"rule_id"`
+	Platform string  `json:"platform"`
+	GroupID  *int64  `json:"group_id,omitempty"`
+	Region   *string `json:"region,omitempty"`
+
+	Until  time.Time `json:"until"`
+	Reason string    `json:"reason"`
+
+	CreatedBy *int64    `json:"created_by,omitempty"`
+	CreatedAt time.Time `json:"created_at"`
+}
+
 type OpsAlertEventFilter struct {
 	Limit int

+	// Cursor pagination (descending by fired_at, then id).
+	BeforeFiredAt *time.Time
+	BeforeID      *int64
+
 	// Optional filters.
-	Status   string
-	Severity string
+	Status    string
+	Severity  string
+	EmailSent *bool

 	StartTime *time.Time
 	EndTime   *time.Time
--- a/backend/internal/service/ops_alerts.go
+++ b/backend/internal/service/ops_alerts.go
@@ -88,6 +88,29 @@ func (s *OpsService) ListAlertEvents(ctx context.Context, filter *OpsAlertEventF
 	return s.opsRepo.ListAlertEvents(ctx, filter)
 }

+func (s *OpsService) GetAlertEventByID(ctx context.Context, eventID int64) (*OpsAlertEvent, error) {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return nil, err
+	}
+	if s.opsRepo == nil {
+		return nil, infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if eventID <= 0 {
+		return nil, infraerrors.BadRequest("INVALID_EVENT_ID", "invalid event id")
+	}
+	ev, err := s.opsRepo.GetAlertEventByID(ctx, eventID)
+	if err != nil {
+		if errors.Is(err, sql.ErrNoRows) {
+			return nil, infraerrors.NotFound("OPS_ALERT_EVENT_NOT_FOUND", "alert event not found")
+		}
+		return nil, err
+	}
+	if ev == nil {
+		return nil, infraerrors.NotFound("OPS_ALERT_EVENT_NOT_FOUND", "alert event not found")
+	}
+	return ev, nil
+}
+
 func (s *OpsService) GetActiveAlertEvent(ctx context.Context, ruleID int64) (*OpsAlertEvent, error) {
 	if err := s.RequireMonitoringEnabled(ctx); err != nil {
 		return nil, err
@@ -101,6 +124,49 @@ func (s *OpsService) GetActiveAlertEvent(ctx context.Context, ruleID int64) (*Op
 	return s.opsRepo.GetActiveAlertEvent(ctx, ruleID)
 }

+func (s *OpsService) CreateAlertSilence(ctx context.Context, input *OpsAlertSilence) (*OpsAlertSilence, error) {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return nil, err
+	}
+	if s.opsRepo == nil {
+		return nil, infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if input == nil {
+		return nil, infraerrors.BadRequest("INVALID_SILENCE", "invalid silence")
+	}
+	if input.RuleID <= 0 {
+		return nil, infraerrors.BadRequest("INVALID_RULE_ID", "invalid rule id")
+	}
+	if strings.TrimSpace(input.Platform) == "" {
+		return nil, infraerrors.BadRequest("INVALID_PLATFORM", "invalid platform")
+	}
+	if input.Until.IsZero() {
+		return nil, infraerrors.BadRequest("INVALID_UNTIL", "invalid until")
+	}
+
+	created, err := s.opsRepo.CreateAlertSilence(ctx, input)
+	if err != nil {
+		return nil, err
+	}
+	return created, nil
+}
+
+func (s *OpsService) IsAlertSilenced(ctx context.Context, ruleID int64, platform string, groupID *int64, region *string, now time.Time) (bool, error) {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return false, err
+	}
+	if s.opsRepo == nil {
+		return false, infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if ruleID <= 0 {
+		return false, infraerrors.BadRequest("INVALID_RULE_ID", "invalid rule id")
+	}
+	if strings.TrimSpace(platform) == "" {
+		return false, nil
+	}
+	return s.opsRepo.IsAlertSilenced(ctx, ruleID, platform, groupID, region, now)
+}
+
 func (s *OpsService) GetLatestAlertEvent(ctx context.Context, ruleID int64) (*OpsAlertEvent, error) {
 	if err := s.RequireMonitoringEnabled(ctx); err != nil {
 		return nil, err
@@ -142,7 +208,11 @@ func (s *OpsService) UpdateAlertEventStatus(ctx context.Context, eventID int64,
 	if eventID <= 0 {
 		return infraerrors.BadRequest("INVALID_EVENT_ID", "invalid event id")
 	}
-	if strings.TrimSpace(status) == "" {
+	status = strings.TrimSpace(status)
+	if status == "" {
+		return infraerrors.BadRequest("INVALID_STATUS", "invalid status")
+	}
+	if status != OpsAlertStatusResolved && status != OpsAlertStatusManualResolved {
 		return infraerrors.BadRequest("INVALID_STATUS", "invalid status")
 	}
 	return s.opsRepo.UpdateAlertEventStatus(ctx, eventID, status, resolvedAt)
--- a/backend/internal/service/ops_health_score.go
+++ b/backend/internal/service/ops_health_score.go
@@ -32,49 +32,38 @@ func computeDashboardHealthScore(now time.Time, overview *OpsDashboardOverview)
 }

 // computeBusinessHealth calculates business health score (0-100)
-// Components: SLA (50%) + Error Rate (30%) + Latency (20%)
+// Components: Error Rate (50%) + TTFT (50%)
 func computeBusinessHealth(overview *OpsDashboardOverview) float64 {
-	// SLA score: 99.5% → 100, 95% → 0 (linear)
-	slaScore := 100.0
-	slaPct := clampFloat64(overview.SLA*100, 0, 100)
-	if slaPct < 99.5 {
-		if slaPct >= 95 {
-			slaScore = (slaPct - 95) / 4.5 * 100
-		} else {
-			slaScore = 0
-		}
-	}
-
-	// Error rate score: 0.5% → 100, 5% → 0 (linear)
+	// Error rate score: 1% → 100, 10% → 0 (linear)
 	// Combines request errors and upstream errors
 	errorScore := 100.0
 	errorPct := clampFloat64(overview.ErrorRate*100, 0, 100)
 	upstreamPct := clampFloat64(overview.UpstreamErrorRate*100, 0, 100)
 	combinedErrorPct := math.Max(errorPct, upstreamPct) // Use worst case
-	if combinedErrorPct > 0.5 {
-		if combinedErrorPct <= 5 {
-			errorScore = (5 - combinedErrorPct) / 4.5 * 100
+	if combinedErrorPct > 1.0 {
+		if combinedErrorPct <= 10.0 {
+			errorScore = (10.0 - combinedErrorPct) / 9.0 * 100
 		} else {
 			errorScore = 0
 		}
 	}

-	// Latency score: 1s → 100, 10s → 0 (linear)
-	// Uses P99 of duration (TTFT is less critical for overall health)
-	latencyScore := 100.0
-	if overview.Duration.P99 != nil {
-		p99 := float64(*overview.Duration.P99)
+	// TTFT score: 1s → 100, 3s → 0 (linear)
+	// Time to first token is critical for user experience
+	ttftScore := 100.0
+	if overview.TTFT.P99 != nil {
+		p99 := float64(*overview.TTFT.P99)
 		if p99 > 1000 {
-			if p99 <= 10000 {
-				latencyScore = (10000 - p99) / 9000 * 100
+			if p99 <= 3000 {
+				ttftScore = (3000 - p99) / 2000 * 100
 			} else {
-				latencyScore = 0
+				ttftScore = 0
 			}
 		}
 	}

-	// Weighted combination
-	return slaScore*0.5 + errorScore*0.3 + latencyScore*0.2
+	// Weighted combination: 50% error rate + 50% TTFT
+	return errorScore*0.5 + ttftScore*0.5
 }

 // computeInfraHealth calculates infrastructure health score (0-100)
--- a/backend/internal/service/ops_health_score_test.go
+++ b/backend/internal/service/ops_health_score_test.go
@@ -127,8 +127,8 @@ func TestComputeDashboardHealthScore_Comprehensive(t *testing.T) {
 					MemoryUsagePercent: float64Ptr(75),
 				},
 			},
-			wantMin: 60,
-			wantMax: 85,
+			wantMin: 96,
+			wantMax: 97,
 		},
 		{
 			name: "DB failure",
@@ -203,8 +203,8 @@ func TestComputeDashboardHealthScore_Comprehensive(t *testing.T) {
 					MemoryUsagePercent: float64Ptr(30),
 				},
 			},
-			wantMin: 25,
-			wantMax: 50,
+			wantMin: 84,
+			wantMax: 85,
 		},
 		{
 			name: "combined failures - business healthy + infra degraded",
@@ -277,30 +277,41 @@ func TestComputeBusinessHealth(t *testing.T) {
 				UpstreamErrorRate: 0,
 				Duration:          OpsPercentiles{P99: intPtr(500)},
 			},
-			wantMin: 50,
-			wantMax: 60,
+			wantMin: 100,
+			wantMax: 100,
 		},
 		{
-			name: "error rate boundary 0.5%",
+			name: "error rate boundary 1%",
 			overview: &OpsDashboardOverview{
-				SLA:               0.995,
-				ErrorRate:         0.005,
+				SLA:               0.99,
+				ErrorRate:         0.01,
 				UpstreamErrorRate: 0,
 				Duration:          OpsPercentiles{P99: intPtr(500)},
 			},
-			wantMin: 95,
+			wantMin: 100,
 			wantMax: 100,
 		},
 		{
-			name: "latency boundary 1000ms",
+			name: "error rate 5%",
 			overview: &OpsDashboardOverview{
-				SLA:               0.995,
+				SLA:               0.95,
+				ErrorRate:         0.05,
+				UpstreamErrorRate: 0,
+				Duration:          OpsPercentiles{P99: intPtr(500)},
+			},
+			wantMin: 77,
+			wantMax: 78,
+		},
+		{
+			name: "TTFT boundary 2s",
+			overview: &OpsDashboardOverview{
+				SLA:               0.99,
 				ErrorRate:         0,
 				UpstreamErrorRate: 0,
-				Duration:          OpsPercentiles{P99: intPtr(1000)},
+				TTFT:              OpsPercentiles{P99: intPtr(2000)},
 			},
-			wantMin: 95,
-			wantMax: 100,
+			wantMin: 75,
+			wantMax: 75,
 		},
 		{
 			name: "upstream error dominates",
@@ -310,7 +321,7 @@ func TestComputeBusinessHealth(t *testing.T) {
 				UpstreamErrorRate: 0.03,
 				Duration:          OpsPercentiles{P99: intPtr(500)},
 			},
-			wantMin: 75,
+			wantMin: 88,
 			wantMax: 90,
 		},
 	}
--- a/backend/internal/service/ops_models.go
+++ b/backend/internal/service/ops_models.go
@@ -6,24 +6,43 @@ type OpsErrorLog struct {
 	ID        int64     `json:"id"`
 	CreatedAt time.Time `json:"created_at"`

-	Phase    string `json:"phase"`
-	Type     string `json:"type"`
+	// Standardized classification
+	// - phase: request|auth|routing|upstream|network|internal
+	// - owner: client|provider|platform
+	// - source: client_request|upstream_http|gateway
+	Phase string `json:"phase"`
+	Type  string `json:"type"`
+
+	Owner  string `json:"error_owner"`
+	Source string `json:"error_source"`
+
 	Severity string `json:"severity"`

 	StatusCode int    `json:"status_code"`
 	Platform   string `json:"platform"`
 	Model      string `json:"model"`

-	LatencyMs *int `json:"latency_ms"`
+	IsRetryable bool `json:"is_retryable"`
+	RetryCount  int  `json:"retry_count"`
+
+	Resolved           bool       `json:"resolved"`
+	ResolvedAt         *time.Time `json:"resolved_at"`
+	ResolvedByUserID   *int64     `json:"resolved_by_user_id"`
+	ResolvedByUserName string     `json:"resolved_by_user_name"`
+	ResolvedRetryID    *int64     `json:"resolved_retry_id"`
+	ResolvedStatusRaw  string     `json:"-"`

 	ClientRequestID string `json:"client_request_id"`
 	RequestID       string `json:"request_id"`
 	Message         string `json:"message"`

-	UserID    *int64 `json:"user_id"`
-	APIKeyID  *int64 `json:"api_key_id"`
-	AccountID *int64 `json:"account_id"`
-	GroupID   *int64 `json:"group_id"`
+	UserID      *int64 `json:"user_id"`
+	UserEmail   string `json:"user_email"`
+	APIKeyID    *int64 `json:"api_key_id"`
+	AccountID   *int64 `json:"account_id"`
+	AccountName string `json:"account_name"`
+	GroupID     *int64 `json:"group_id"`
+	GroupName   string `json:"group_name"`

 	ClientIP    *string `json:"client_ip"`
 	RequestPath string  `json:"request_path"`
@@ -67,9 +86,24 @@ type OpsErrorLogFilter struct {
 	GroupID   *int64
 	AccountID *int64

-	StatusCodes []int
-	Phase       string
-	Query       string
+	StatusCodes      []int
+	StatusCodesOther bool
+	Phase            string
+	Owner            string
+	Source           string
+	Resolved         *bool
+	Query            string
+	UserQuery        string // Search by user email
+
+	// Optional correlation keys for exact matching.
+	RequestID       string
+	ClientRequestID string
+
+	// View controls error categorization for list endpoints.
+	// - errors: show actionable errors (exclude business-limited / 429 / 529)
+	// - excluded: only show excluded errors
+	// - all: show everything
+	View string

 	Page     int
 	PageSize int
@@ -90,12 +124,23 @@ type OpsRetryAttempt struct {
 	SourceErrorID     int64  `json:"source_error_id"`
 	Mode              string `json:"mode"`
 	PinnedAccountID   *int64 `json:"pinned_account_id"`
+	PinnedAccountName string `json:"pinned_account_name"`

 	Status     string     `json:"status"`
 	StartedAt  *time.Time `json:"started_at"`
 	FinishedAt *time.Time `json:"finished_at"`
 	DurationMs *int64     `json:"duration_ms"`

+	// Persisted execution results (best-effort)
+	Success           *bool   `json:"success"`
+	HTTPStatusCode    *int    `json:"http_status_code"`
+	UpstreamRequestID *string `json:"upstream_request_id"`
+	UsedAccountID     *int64  `json:"used_account_id"`
+	UsedAccountName   string  `json:"used_account_name"`
+	ResponsePreview   *string `json:"response_preview"`
+	ResponseTruncated *bool   `json:"response_truncated"`
+
+	// Optional correlation
 	ResultRequestID *string `json:"result_request_id"`
 	ResultErrorID   *int64  `json:"result_error_id"`

--- a/backend/internal/service/ops_port.go
+++ b/backend/internal/service/ops_port.go
@@ -14,6 +14,8 @@ type OpsRepository interface {
 	InsertRetryAttempt(ctx context.Context, input *OpsInsertRetryAttemptInput) (int64, error)
 	UpdateRetryAttempt(ctx context.Context, input *OpsUpdateRetryAttemptInput) error
 	GetLatestRetryAttemptForError(ctx context.Context, sourceErrorID int64) (*OpsRetryAttempt, error)
+	ListRetryAttemptsByErrorID(ctx context.Context, sourceErrorID int64, limit int) ([]*OpsRetryAttempt, error)
+	UpdateErrorResolution(ctx context.Context, errorID int64, resolved bool, resolvedByUserID *int64, resolvedRetryID *int64, resolvedAt *time.Time) error

 	// Lightweight window stats (for realtime WS / quick sampling).
 	GetWindowStats(ctx context.Context, filter *OpsDashboardFilter) (*OpsWindowStats, error)
@@ -39,12 +41,17 @@ type OpsRepository interface {
 	DeleteAlertRule(ctx context.Context, id int64) error

 	ListAlertEvents(ctx context.Context, filter *OpsAlertEventFilter) ([]*OpsAlertEvent, error)
+	GetAlertEventByID(ctx context.Context, eventID int64) (*OpsAlertEvent, error)
 	GetActiveAlertEvent(ctx context.Context, ruleID int64) (*OpsAlertEvent, error)
 	GetLatestAlertEvent(ctx context.Context, ruleID int64) (*OpsAlertEvent, error)
 	CreateAlertEvent(ctx context.Context, event *OpsAlertEvent) (*OpsAlertEvent, error)
 	UpdateAlertEventStatus(ctx context.Context, eventID int64, status string, resolvedAt *time.Time) error
 	UpdateAlertEventEmailSent(ctx context.Context, eventID int64, emailSent bool) error

+	// Alert silences
+	CreateAlertSilence(ctx context.Context, input *OpsAlertSilence) (*OpsAlertSilence, error)
+	IsAlertSilenced(ctx context.Context, ruleID int64, platform string, groupID *int64, region *string, now time.Time) (bool, error)
+
 	// Pre-aggregation (hourly/daily) used for long-window dashboard performance.
 	UpsertHourlyMetrics(ctx context.Context, startTime, endTime time.Time) error
 	UpsertDailyMetrics(ctx context.Context, startTime, endTime time.Time) error
@@ -91,7 +98,6 @@ type OpsInsertErrorLogInput struct {
 	// It is set by OpsService.RecordError before persisting.
 	UpstreamErrorsJSON *string

-	DurationMs         *int
 	TimeToFirstTokenMs *int64

 	RequestBodyJSON      *string // sanitized json string (not raw bytes)
@@ -124,7 +130,15 @@ type OpsUpdateRetryAttemptInput struct {
 	FinishedAt time.Time
 	DurationMs int64

-	// Optional correlation
+	// Persisted execution results (best-effort)
+	Success           *bool
+	HTTPStatusCode    *int
+	UpstreamRequestID *string
+	UsedAccountID     *int64
+	ResponsePreview   *string
+	ResponseTruncated *bool
+
+	// Optional correlation (legacy fields kept)
 	ResultRequestID *string
 	ResultErrorID   *int64

--- a/backend/internal/service/ops_retry.go
+++ b/backend/internal/service/ops_retry.go
@@ -108,6 +108,10 @@ func (w *limitedResponseWriter) truncated() bool {
 	return w.totalWritten > int64(w.limit)
 }

+const (
+	OpsRetryModeUpstreamEvent = "upstream_event"
+)
+
 func (s *OpsService) RetryError(ctx context.Context, requestedByUserID int64, errorID int64, mode string, pinnedAccountID *int64) (*OpsRetryResult, error) {
 	if err := s.RequireMonitoringEnabled(ctx); err != nil {
 		return nil, err
@@ -123,6 +127,81 @@ func (s *OpsService) RetryError(ctx context.Context, requestedByUserID int64, er
 		return nil, infraerrors.BadRequest("OPS_RETRY_INVALID_MODE", "mode must be client or upstream")
 	}

+	errorLog, err := s.GetErrorLogByID(ctx, errorID)
+	if err != nil {
+		return nil, err
+	}
+	if errorLog == nil {
+		return nil, infraerrors.NotFound("OPS_ERROR_NOT_FOUND", "ops error log not found")
+	}
+	if strings.TrimSpace(errorLog.RequestBody) == "" {
+		return nil, infraerrors.BadRequest("OPS_RETRY_NO_REQUEST_BODY", "No request body found to retry")
+	}
+
+	var pinned *int64
+	if mode == OpsRetryModeUpstream {
+		if pinnedAccountID != nil && *pinnedAccountID > 0 {
+			pinned = pinnedAccountID
+		} else if errorLog.AccountID != nil && *errorLog.AccountID > 0 {
+			pinned = errorLog.AccountID
+		} else {
+			return nil, infraerrors.BadRequest("OPS_RETRY_PINNED_ACCOUNT_REQUIRED", "pinned_account_id is required for upstream retry")
+		}
+	}
+
+	return s.retryWithErrorLog(ctx, requestedByUserID, errorID, mode, mode, pinned, errorLog)
+}
+
+// RetryUpstreamEvent retries a specific upstream attempt captured inside ops_error_logs.upstream_errors.
+// idx is 0-based. It always pins the original event account_id.
+func (s *OpsService) RetryUpstreamEvent(ctx context.Context, requestedByUserID int64, errorID int64, idx int) (*OpsRetryResult, error) {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return nil, err
+	}
+	if s.opsRepo == nil {
+		return nil, infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if idx < 0 {
+		return nil, infraerrors.BadRequest("OPS_RETRY_INVALID_UPSTREAM_IDX", "invalid upstream idx")
+	}
+
+	errorLog, err := s.GetErrorLogByID(ctx, errorID)
+	if err != nil {
+		return nil, err
+	}
+	if errorLog == nil {
+		return nil, infraerrors.NotFound("OPS_ERROR_NOT_FOUND", "ops error log not found")
+	}
+
+	events, err := ParseOpsUpstreamErrors(errorLog.UpstreamErrors)
+	if err != nil {
+		return nil, infraerrors.BadRequest("OPS_RETRY_UPSTREAM_EVENTS_INVALID", "invalid upstream_errors")
+	}
+	if idx >= len(events) {
+		return nil, infraerrors.BadRequest("OPS_RETRY_UPSTREAM_IDX_OOB", "upstream idx out of range")
+	}
+	ev := events[idx]
+	if ev == nil {
+		return nil, infraerrors.BadRequest("OPS_RETRY_UPSTREAM_EVENT_MISSING", "upstream event missing")
+	}
+	if ev.AccountID <= 0 {
+		return nil, infraerrors.BadRequest("OPS_RETRY_PINNED_ACCOUNT_REQUIRED", "account_id is required for upstream retry")
+	}
+
+	upstreamBody := strings.TrimSpace(ev.UpstreamRequestBody)
+	if upstreamBody == "" {
+		return nil, infraerrors.BadRequest("OPS_RETRY_UPSTREAM_NO_REQUEST_BODY", "No upstream request body found to retry")
+	}
+
+	override := *errorLog
+	override.RequestBody = upstreamBody
+	pinned := ev.AccountID
+
+	// Persist as upstream_event, execute as upstream pinned retry.
+	return s.retryWithErrorLog(ctx, requestedByUserID, errorID, OpsRetryModeUpstreamEvent, OpsRetryModeUpstream, &pinned, &override)
+}
+
+func (s *OpsService) retryWithErrorLog(ctx context.Context, requestedByUserID int64, errorID int64, mode string, execMode string, pinnedAccountID *int64, errorLog *OpsErrorLogDetail) (*OpsRetryResult, error) {
 	latest, err := s.opsRepo.GetLatestRetryAttemptForError(ctx, errorID)
 	if err != nil && !errors.Is(err, sql.ErrNoRows) {
 		return nil, infraerrors.InternalServer("OPS_RETRY_LOAD_LATEST_FAILED", "Failed to check retry status").WithCause(err)
@@ -144,22 +223,18 @@ func (s *OpsService) RetryError(ctx context.Context, requestedByUserID int64, er
 		}
 	}

-	errorLog, err := s.GetErrorLogByID(ctx, errorID)
-	if err != nil {
-		return nil, err
-	}
-	if strings.TrimSpace(errorLog.RequestBody) == "" {
+	if errorLog == nil || strings.TrimSpace(errorLog.RequestBody) == "" {
 		return nil, infraerrors.BadRequest("OPS_RETRY_NO_REQUEST_BODY", "No request body found to retry")
 	}

 	var pinned *int64
-	if mode == OpsRetryModeUpstream {
+	if execMode == OpsRetryModeUpstream {
 		if pinnedAccountID != nil && *pinnedAccountID > 0 {
 			pinned = pinnedAccountID
 		} else if errorLog.AccountID != nil && *errorLog.AccountID > 0 {
 			pinned = errorLog.AccountID
 		} else {
-			return nil, infraerrors.BadRequest("OPS_RETRY_PINNED_ACCOUNT_REQUIRED", "pinned_account_id is required for upstream retry")
+			return nil, infraerrors.BadRequest("OPS_RETRY_PINNED_ACCOUNT_REQUIRED", "account_id is required for upstream retry")
 		}
 	}

@@ -196,7 +271,7 @@ func (s *OpsService) RetryError(ctx context.Context, requestedByUserID int64, er
 	execCtx, cancel := context.WithTimeout(ctx, opsRetryTimeout)
 	defer cancel()

-	execRes := s.executeRetry(execCtx, errorLog, mode, pinned)
+	execRes := s.executeRetry(execCtx, errorLog, execMode, pinned)

 	finishedAt := time.Now()
 	result.FinishedAt = finishedAt
@@ -220,27 +295,40 @@ func (s *OpsService) RetryError(ctx context.Context, requestedByUserID int64, er
 		msg := result.ErrorMessage
 		updateErrMsg = &msg
 	}
+	// Keep legacy result_request_id empty; use upstream_request_id instead.
 	var resultRequestID *string
-	if strings.TrimSpace(result.UpstreamRequestID) != "" {
-		v := result.UpstreamRequestID
-		resultRequestID = &v
-	}

 	finalStatus := result.Status
 	if strings.TrimSpace(finalStatus) == "" {
 		finalStatus = opsRetryStatusFailed
 	}

+	success := strings.EqualFold(finalStatus, opsRetryStatusSucceeded)
+	httpStatus := result.HTTPStatusCode
+	upstreamReqID := result.UpstreamRequestID
+	usedAccountID := result.UsedAccountID
+	preview := result.ResponsePreview
+	truncated := result.ResponseTruncated
+
 	if err := s.opsRepo.UpdateRetryAttempt(updateCtx, &OpsUpdateRetryAttemptInput{
-		ID:              attemptID,
-		Status:          finalStatus,
-		FinishedAt:      finishedAt,
-		DurationMs:      result.DurationMs,
-		ResultRequestID: resultRequestID,
-		ErrorMessage:    updateErrMsg,
+		ID:                attemptID,
+		Status:            finalStatus,
+		FinishedAt:        finishedAt,
+		DurationMs:        result.DurationMs,
+		Success:           &success,
+		HTTPStatusCode:    &httpStatus,
+		UpstreamRequestID: &upstreamReqID,
+		UsedAccountID:     usedAccountID,
+		ResponsePreview:   &preview,
+		ResponseTruncated: &truncated,
+		ResultRequestID:   resultRequestID,
+		ErrorMessage:      updateErrMsg,
 	}); err != nil {
-		// Best-effort: retry itself already executed; do not fail the API response.
 		log.Printf("[Ops] UpdateRetryAttempt failed: %v", err)
+	} else if success {
+		if err := s.opsRepo.UpdateErrorResolution(updateCtx, errorID, true, &requestedByUserID, &attemptID, &finishedAt); err != nil {
+			log.Printf("[Ops] UpdateErrorResolution failed: %v", err)
+		}
 	}

 	return result, nil
--- a/backend/internal/service/ops_service.go
+++ b/backend/internal/service/ops_service.go
@@ -208,6 +208,25 @@ func (s *OpsService) RecordError(ctx context.Context, entry *OpsInsertErrorLogIn
 				out.Detail = ""
 			}

+			out.UpstreamRequestBody = strings.TrimSpace(out.UpstreamRequestBody)
+			if out.UpstreamRequestBody != "" {
+				// Reuse the same sanitization/trimming strategy as request body storage.
+				// Keep it small so it is safe to persist in ops_error_logs JSON.
+				sanitized, truncated, _ := sanitizeAndTrimRequestBody([]byte(out.UpstreamRequestBody), 10*1024)
+				if sanitized != "" {
+					out.UpstreamRequestBody = sanitized
+					if truncated {
+						out.Kind = strings.TrimSpace(out.Kind)
+						if out.Kind == "" {
+							out.Kind = "upstream"
+						}
+						out.Kind = out.Kind + ":request_body_truncated"
+					}
+				} else {
+					out.UpstreamRequestBody = ""
+				}
+			}
+
 			// Drop fully-empty events (can happen if only status code was known).
 			if out.UpstreamStatusCode == 0 && out.Message == "" && out.Detail == "" {
 				continue
@@ -236,7 +255,13 @@ func (s *OpsService) GetErrorLogs(ctx context.Context, filter *OpsErrorLogFilter
 	if s.opsRepo == nil {
 		return &OpsErrorLogList{Errors: []*OpsErrorLog{}, Total: 0, Page: 1, PageSize: 20}, nil
 	}
-	return s.opsRepo.ListErrorLogs(ctx, filter)
+	result, err := s.opsRepo.ListErrorLogs(ctx, filter)
+	if err != nil {
+		log.Printf("[Ops] GetErrorLogs failed: %v", err)
+		return nil, err
+	}
+
+	return result, nil
 }

 func (s *OpsService) GetErrorLogByID(ctx context.Context, id int64) (*OpsErrorLogDetail, error) {
@@ -256,6 +281,46 @@ func (s *OpsService) GetErrorLogByID(ctx context.Context, id int64) (*OpsErrorLo
 	return detail, nil
 }

+func (s *OpsService) ListRetryAttemptsByErrorID(ctx context.Context, errorID int64, limit int) ([]*OpsRetryAttempt, error) {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return nil, err
+	}
+	if s.opsRepo == nil {
+		return nil, infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if errorID <= 0 {
+		return nil, infraerrors.BadRequest("OPS_ERROR_INVALID_ID", "invalid error id")
+	}
+	items, err := s.opsRepo.ListRetryAttemptsByErrorID(ctx, errorID, limit)
+	if err != nil {
+		if errors.Is(err, sql.ErrNoRows) {
+			return []*OpsRetryAttempt{}, nil
+		}
+		return nil, infraerrors.InternalServer("OPS_RETRY_LIST_FAILED", "Failed to list retry attempts").WithCause(err)
+	}
+	return items, nil
+}
+
+func (s *OpsService) UpdateErrorResolution(ctx context.Context, errorID int64, resolved bool, resolvedByUserID *int64, resolvedRetryID *int64) error {
+	if err := s.RequireMonitoringEnabled(ctx); err != nil {
+		return err
+	}
+	if s.opsRepo == nil {
+		return infraerrors.ServiceUnavailable("OPS_REPO_UNAVAILABLE", "Ops repository not available")
+	}
+	if errorID <= 0 {
+		return infraerrors.BadRequest("OPS_ERROR_INVALID_ID", "invalid error id")
+	}
+	// Best-effort ensure the error exists
+	if _, err := s.opsRepo.GetErrorLogByID(ctx, errorID); err != nil {
+		if errors.Is(err, sql.ErrNoRows) {
+			return infraerrors.NotFound("OPS_ERROR_NOT_FOUND", "ops error log not found")
+		}
+		return infraerrors.InternalServer("OPS_ERROR_LOAD_FAILED", "Failed to load ops error log").WithCause(err)
+	}
+	return s.opsRepo.UpdateErrorResolution(ctx, errorID, resolved, resolvedByUserID, resolvedRetryID, nil)
+}
+
 func sanitizeAndTrimRequestBody(raw []byte, maxBytes int) (jsonString string, truncated bool, bytesLen int) {
 	bytesLen = len(raw)
 	if len(raw) == 0 {
@@ -296,14 +361,34 @@ func sanitizeAndTrimRequestBody(raw []byte, maxBytes int) (jsonString string, tr
 		}
 	}

-	// Last resort: store a minimal placeholder (still valid JSON).
-	placeholder := map[string]any{
-		"request_body_truncated": true,
+	// Last resort: keep JSON shape but drop big fields.
+	// This avoids downstream code that expects certain top-level keys from crashing.
+	if root, ok := decoded.(map[string]any); ok {
+		placeholder := shallowCopyMap(root)
+		placeholder["request_body_truncated"] = true
+
+		// Replace potentially huge arrays/strings, but keep the keys present.
+		for _, k := range []string{"messages", "contents", "input", "prompt"} {
+			if _, exists := placeholder[k]; exists {
+				placeholder[k] = []any{}
+			}
+		}
+		for _, k := range []string{"text"} {
+			if _, exists := placeholder[k]; exists {
+				placeholder[k] = ""
+			}
+		}
+
+		encoded4, err4 := json.Marshal(placeholder)
+		if err4 == nil {
+			if len(encoded4) <= maxBytes {
+				return string(encoded4), true, bytesLen
+			}
+		}
 	}
-	if model := extractString(decoded, "model"); model != "" {
-		placeholder["model"] = model
-	}
-	encoded4, err4 := json.Marshal(placeholder)
+
+	// Final fallback: minimal valid JSON.
+	encoded4, err4 := json.Marshal(map[string]any{"request_body_truncated": true})
 	if err4 != nil {
 		return "", true, bytesLen
 	}
@@ -526,12 +611,3 @@ func sanitizeErrorBodyForStorage(raw string, maxBytes int) (sanitized string, tr
 	}
 	return raw, false
 }
-
-func extractString(v any, key string) string {
-	root, ok := v.(map[string]any)
-	if !ok {
-		return ""
-	}
-	s, _ := root[key].(string)
-	return strings.TrimSpace(s)
-}
--- a/backend/internal/service/ops_settings.go
+++ b/backend/internal/service/ops_settings.go
@@ -368,9 +368,11 @@ func defaultOpsAdvancedSettings() *OpsAdvancedSettings {
 		Aggregation: OpsAggregationSettings{
 			AggregationEnabled: false,
 		},
-		IgnoreCountTokensErrors: false,
-		AutoRefreshEnabled:      false,
-		AutoRefreshIntervalSec:  30,
+		IgnoreCountTokensErrors:   false,
+		IgnoreContextCanceled:     true,  // Default to true - client disconnects are not errors
+		IgnoreNoAvailableAccounts: false, // Default to false - this is a real routing issue
+		AutoRefreshEnabled:        false,
+		AutoRefreshIntervalSec:    30,
 	}
 }

@@ -482,13 +484,11 @@ const SettingKeyOpsMetricThresholds = "ops_metric_thresholds"

 func defaultOpsMetricThresholds() *OpsMetricThresholds {
 	slaMin := 99.5
-	latencyMax := 2000.0
 	ttftMax := 500.0
 	reqErrMax := 5.0
 	upstreamErrMax := 5.0
 	return &OpsMetricThresholds{
 		SLAPercentMin:               &slaMin,
-		LatencyP99MsMax:             &latencyMax,
 		TTFTp99MsMax:                &ttftMax,
 		RequestErrorRatePercentMax:  &reqErrMax,
 		UpstreamErrorRatePercentMax: &upstreamErrMax,
@@ -538,9 +538,6 @@ func (s *OpsService) UpdateMetricThresholds(ctx context.Context, cfg *OpsMetricT
 	if cfg.SLAPercentMin != nil && (*cfg.SLAPercentMin < 0 || *cfg.SLAPercentMin > 100) {
 		return nil, errors.New("sla_percent_min must be between 0 and 100")
 	}
-	if cfg.LatencyP99MsMax != nil && *cfg.LatencyP99MsMax < 0 {
-		return nil, errors.New("latency_p99_ms_max must be >= 0")
-	}
 	if cfg.TTFTp99MsMax != nil && *cfg.TTFTp99MsMax < 0 {
 		return nil, errors.New("ttft_p99_ms_max must be >= 0")
 	}
--- a/backend/internal/service/ops_settings_models.go
+++ b/backend/internal/service/ops_settings_models.go
@@ -63,7 +63,6 @@ type OpsAlertSilencingSettings struct {

 type OpsMetricThresholds struct {
 	SLAPercentMin               *float64 `json:"sla_percent_min,omitempty"`                 // SLA低于此值变红
-	LatencyP99MsMax             *float64 `json:"latency_p99_ms_max,omitempty"`              // 延迟P99高于此值变红
 	TTFTp99MsMax                *float64 `json:"ttft_p99_ms_max,omitempty"`                 // TTFT P99高于此值变红
 	RequestErrorRatePercentMax  *float64 `json:"request_error_rate_percent_max,omitempty"`  // 请求错误率高于此值变红
 	UpstreamErrorRatePercentMax *float64 `json:"upstream_error_rate_percent_max,omitempty"` // 上游错误率高于此值变红
@@ -79,11 +78,13 @@ type OpsAlertRuntimeSettings struct {

 // OpsAdvancedSettings stores advanced ops configuration (data retention, aggregation).
 type OpsAdvancedSettings struct {
-	DataRetention           OpsDataRetentionSettings `json:"data_retention"`
-	Aggregation             OpsAggregationSettings   `json:"aggregation"`
-	IgnoreCountTokensErrors bool                     `json:"ignore_count_tokens_errors"`
-	AutoRefreshEnabled      bool                     `json:"auto_refresh_enabled"`
-	AutoRefreshIntervalSec  int                      `json:"auto_refresh_interval_seconds"`
+	DataRetention             OpsDataRetentionSettings `json:"data_retention"`
+	Aggregation               OpsAggregationSettings   `json:"aggregation"`
+	IgnoreCountTokensErrors   bool                     `json:"ignore_count_tokens_errors"`
+	IgnoreContextCanceled     bool                     `json:"ignore_context_canceled"`
+	IgnoreNoAvailableAccounts bool                     `json:"ignore_no_available_accounts"`
+	AutoRefreshEnabled        bool                     `json:"auto_refresh_enabled"`
+	AutoRefreshIntervalSec    int                      `json:"auto_refresh_interval_seconds"`
 }

 type OpsDataRetentionSettings struct {
--- a/backend/internal/service/ops_upstream_context.go
+++ b/backend/internal/service/ops_upstream_context.go
@@ -15,6 +15,11 @@ const (
 	OpsUpstreamErrorMessageKey = "ops_upstream_error_message"
 	OpsUpstreamErrorDetailKey  = "ops_upstream_error_detail"
 	OpsUpstreamErrorsKey       = "ops_upstream_errors"
+
+	// Best-effort capture of the current upstream request body so ops can
+	// retry the specific upstream attempt (not just the client request).
+	// This value is sanitized+trimmed before being persisted.
+	OpsUpstreamRequestBodyKey = "ops_upstream_request_body"
 )

 func setOpsUpstreamError(c *gin.Context, upstreamStatusCode int, upstreamMessage, upstreamDetail string) {
@@ -38,13 +43,21 @@ type OpsUpstreamErrorEvent struct {
 	AtUnixMs int64 `json:"at_unix_ms,omitempty"`

 	// Context
-	Platform  string `json:"platform,omitempty"`
-	AccountID int64  `json:"account_id,omitempty"`
+	Platform    string `json:"platform,omitempty"`
+	AccountID   int64  `json:"account_id,omitempty"`
+	AccountName string `json:"account_name,omitempty"`

 	// Outcome
 	UpstreamStatusCode int    `json:"upstream_status_code,omitempty"`
 	UpstreamRequestID  string `json:"upstream_request_id,omitempty"`

+	// Best-effort upstream request capture (sanitized+trimmed).
+	// Required for retrying a specific upstream attempt.
+	UpstreamRequestBody string `json:"upstream_request_body,omitempty"`
+
+	// Best-effort upstream response capture (sanitized+trimmed).
+	UpstreamResponseBody string `json:"upstream_response_body,omitempty"`
+
 	// Kind: http_error | request_error | retry_exhausted | failover
 	Kind string `json:"kind,omitempty"`

@@ -61,6 +74,8 @@ func appendOpsUpstreamError(c *gin.Context, ev OpsUpstreamErrorEvent) {
 	}
 	ev.Platform = strings.TrimSpace(ev.Platform)
 	ev.UpstreamRequestID = strings.TrimSpace(ev.UpstreamRequestID)
+	ev.UpstreamRequestBody = strings.TrimSpace(ev.UpstreamRequestBody)
+	ev.UpstreamResponseBody = strings.TrimSpace(ev.UpstreamResponseBody)
 	ev.Kind = strings.TrimSpace(ev.Kind)
 	ev.Message = strings.TrimSpace(ev.Message)
 	ev.Detail = strings.TrimSpace(ev.Detail)
@@ -68,6 +83,16 @@ func appendOpsUpstreamError(c *gin.Context, ev OpsUpstreamErrorEvent) {
 		ev.Message = sanitizeUpstreamErrorMessage(ev.Message)
 	}

+	// If the caller didn't explicitly pass upstream request body but the gateway
+	// stored it on the context, attach it so ops can retry this specific attempt.
+	if ev.UpstreamRequestBody == "" {
+		if v, ok := c.Get(OpsUpstreamRequestBodyKey); ok {
+			if s, ok := v.(string); ok {
+				ev.UpstreamRequestBody = strings.TrimSpace(s)
+			}
+		}
+	}
+
 	var existing []*OpsUpstreamErrorEvent
 	if v, ok := c.Get(OpsUpstreamErrorsKey); ok {
 		if arr, ok := v.([]*OpsUpstreamErrorEvent); ok {
@@ -92,3 +117,15 @@ func marshalOpsUpstreamErrors(events []*OpsUpstreamErrorEvent) *string {
 	s := string(raw)
 	return &s
 }
+
+func ParseOpsUpstreamErrors(raw string) ([]*OpsUpstreamErrorEvent, error) {
+	raw = strings.TrimSpace(raw)
+	if raw == "" {
+		return []*OpsUpstreamErrorEvent{}, nil
+	}
+	var out []*OpsUpstreamErrorEvent
+	if err := json.Unmarshal([]byte(raw), &out); err != nil {
+		return nil, err
+	}
+	return out, nil
+}