feat(ops): 添加ops handler和路由配置

backend/internal/handler/admin/ops_handler.go

@@ -19,6 +19,29 @@ type OpsHandler struct {
 	opsService *service.OpsService
 }
 
+const (
+	opsListViewErrors   = "errors"
+	opsListViewExcluded = "excluded"
+	opsListViewAll      = "all"
+)
+
+func parseOpsViewParam(c *gin.Context) string {
+	if c == nil {
+		return ""
+	}
+	v := strings.ToLower(strings.TrimSpace(c.Query("view")))
+	switch v {
+	case "", opsListViewErrors:
+		return opsListViewErrors
+	case opsListViewExcluded:
+		return opsListViewExcluded
+	case opsListViewAll:
+		return opsListViewAll
+	default:
+		return opsListViewErrors
+	}
+}
+
 func NewOpsHandler(opsService *service.OpsService) *OpsHandler {
 	return &OpsHandler{opsService: opsService}
 }
@@ -86,6 +109,7 @@ func (h *OpsHandler) GetErrorLogs(c *gin.Context) {
 	if source := strings.TrimSpace(c.Query("error_source")); source != "" {
 		filter.Source = source
 	}
+	filter.View = parseOpsViewParam(c)
 	if v := strings.TrimSpace(c.Query("resolved")); v != "" {
 		switch strings.ToLower(v) {
 		case "1", "true", "yes":
@@ -157,6 +181,336 @@ func (h *OpsHandler) GetErrorLogByID(c *gin.Context) {
 	response.Success(c, detail)
 }
 
+// ==================== New split endpoints ====================
+
+// ListRequestErrors lists client-visible request errors.
+// GET /api/v1/admin/ops/request-errors
+func (h *OpsHandler) ListRequestErrors(c *gin.Context) {
+	if h.opsService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
+		return
+	}
+	if err := h.opsService.RequireMonitoringEnabled(c.Request.Context()); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	page, pageSize := response.ParsePagination(c)
+	if pageSize > 500 {
+		pageSize = 500
+	}
+	startTime, endTime, err := parseOpsTimeRange(c, "1h")
+	if err != nil {
+		response.BadRequest(c, err.Error())
+		return
+	}
+
+	filter := &service.OpsErrorLogFilter{Page: page, PageSize: pageSize}
+	if !startTime.IsZero() {
+		filter.StartTime = &startTime
+	}
+	if !endTime.IsZero() {
+		filter.EndTime = &endTime
+	}
+	filter.View = parseOpsViewParam(c)
+	filter.Phase = strings.TrimSpace(c.Query("phase"))
+	filter.Owner = strings.TrimSpace(c.Query("error_owner"))
+	filter.Source = strings.TrimSpace(c.Query("error_source"))
+	filter.Query = strings.TrimSpace(c.Query("q"))
+
+	// Force request errors: client-visible status >= 400.
+	// buildOpsErrorLogsWhere already applies this for non-upstream phase.
+	if strings.EqualFold(strings.TrimSpace(filter.Phase), "upstream") {
+		filter.Phase = ""
+	}
+
+	if platform := strings.TrimSpace(c.Query("platform")); platform != "" {
+		filter.Platform = platform
+	}
+	if v := strings.TrimSpace(c.Query("group_id")); v != "" {
+		id, err := strconv.ParseInt(v, 10, 64)
+		if err != nil || id <= 0 {
+			response.BadRequest(c, "Invalid group_id")
+			return
+		}
+		filter.GroupID = &id
+	}
+	if v := strings.TrimSpace(c.Query("account_id")); v != "" {
+		id, err := strconv.ParseInt(v, 10, 64)
+		if err != nil || id <= 0 {
+			response.BadRequest(c, "Invalid account_id")
+			return
+		}
+		filter.AccountID = &id
+	}
+
+	if v := strings.TrimSpace(c.Query("resolved")); v != "" {
+		switch strings.ToLower(v) {
+		case "1", "true", "yes":
+			b := true
+			filter.Resolved = &b
+		case "0", "false", "no":
+			b := false
+			filter.Resolved = &b
+		default:
+			response.BadRequest(c, "Invalid resolved")
+			return
+		}
+	}
+	if statusCodesStr := strings.TrimSpace(c.Query("status_codes")); statusCodesStr != "" {
+		parts := strings.Split(statusCodesStr, ",")
+		out := make([]int, 0, len(parts))
+		for _, part := range parts {
+			p := strings.TrimSpace(part)
+			if p == "" {
+				continue
+			}
+			n, err := strconv.Atoi(p)
+			if err != nil || n < 0 {
+				response.BadRequest(c, "Invalid status_codes")
+				return
+			}
+			out = append(out, n)
+		}
+		filter.StatusCodes = out
+	}
+
+	result, err := h.opsService.GetErrorLogs(c.Request.Context(), filter)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Paginated(c, result.Errors, int64(result.Total), result.Page, result.PageSize)
+}
+
+// GetRequestError returns request error detail.
+// GET /api/v1/admin/ops/request-errors/:id
+func (h *OpsHandler) GetRequestError(c *gin.Context) {
+	// same storage; just proxy to existing detail
+	h.GetErrorLogByID(c)
+}
+
+// RetryRequestErrorClient retries the client request based on stored request body.
+// POST /api/v1/admin/ops/request-errors/:id/retry-client
+func (h *OpsHandler) RetryRequestErrorClient(c *gin.Context) {
+	if h.opsService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
+		return
+	}
+	if err := h.opsService.RequireMonitoringEnabled(c.Request.Context()); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	subject, ok := middleware.GetAuthSubjectFromContext(c)
+	if !ok || subject.UserID <= 0 {
+		response.Error(c, http.StatusUnauthorized, "Unauthorized")
+		return
+	}
+
+	idStr := strings.TrimSpace(c.Param("id"))
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil || id <= 0 {
+		response.BadRequest(c, "Invalid error id")
+		return
+	}
+
+	result, err := h.opsService.RetryError(c.Request.Context(), subject.UserID, id, service.OpsRetryModeClient, nil)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
+
+// RetryRequestErrorUpstreamEvent retries a specific upstream attempt using captured upstream_request_body.
+// POST /api/v1/admin/ops/request-errors/:id/upstream-errors/:idx/retry
+func (h *OpsHandler) RetryRequestErrorUpstreamEvent(c *gin.Context) {
+	if h.opsService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
+		return
+	}
+	if err := h.opsService.RequireMonitoringEnabled(c.Request.Context()); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	subject, ok := middleware.GetAuthSubjectFromContext(c)
+	if !ok || subject.UserID <= 0 {
+		response.Error(c, http.StatusUnauthorized, "Unauthorized")
+		return
+	}
+
+	idStr := strings.TrimSpace(c.Param("id"))
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil || id <= 0 {
+		response.BadRequest(c, "Invalid error id")
+		return
+	}
+
+	idxStr := strings.TrimSpace(c.Param("idx"))
+	idx, err := strconv.Atoi(idxStr)
+	if err != nil || idx < 0 {
+		response.BadRequest(c, "Invalid upstream idx")
+		return
+	}
+
+	result, err := h.opsService.RetryUpstreamEvent(c.Request.Context(), subject.UserID, id, idx)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
+
+// ResolveRequestError toggles resolved status.
+// PUT /api/v1/admin/ops/request-errors/:id/resolve
+func (h *OpsHandler) ResolveRequestError(c *gin.Context) {
+	h.UpdateErrorResolution(c)
+}
+
+// ListUpstreamErrors lists independent upstream errors.
+// GET /api/v1/admin/ops/upstream-errors
+func (h *OpsHandler) ListUpstreamErrors(c *gin.Context) {
+	if h.opsService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
+		return
+	}
+	if err := h.opsService.RequireMonitoringEnabled(c.Request.Context()); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	page, pageSize := response.ParsePagination(c)
+	if pageSize > 500 {
+		pageSize = 500
+	}
+	startTime, endTime, err := parseOpsTimeRange(c, "1h")
+	if err != nil {
+		response.BadRequest(c, err.Error())
+		return
+	}
+
+	filter := &service.OpsErrorLogFilter{Page: page, PageSize: pageSize}
+	if !startTime.IsZero() {
+		filter.StartTime = &startTime
+	}
+	if !endTime.IsZero() {
+		filter.EndTime = &endTime
+	}
+
+	filter.View = parseOpsViewParam(c)
+	filter.Phase = "upstream"
+	filter.Owner = "provider"
+	filter.Source = strings.TrimSpace(c.Query("error_source"))
+	filter.Query = strings.TrimSpace(c.Query("q"))
+
+	if platform := strings.TrimSpace(c.Query("platform")); platform != "" {
+		filter.Platform = platform
+	}
+	if v := strings.TrimSpace(c.Query("group_id")); v != "" {
+		id, err := strconv.ParseInt(v, 10, 64)
+		if err != nil || id <= 0 {
+			response.BadRequest(c, "Invalid group_id")
+			return
+		}
+		filter.GroupID = &id
+	}
+	if v := strings.TrimSpace(c.Query("account_id")); v != "" {
+		id, err := strconv.ParseInt(v, 10, 64)
+		if err != nil || id <= 0 {
+			response.BadRequest(c, "Invalid account_id")
+			return
+		}
+		filter.AccountID = &id
+	}
+
+	if v := strings.TrimSpace(c.Query("resolved")); v != "" {
+		switch strings.ToLower(v) {
+		case "1", "true", "yes":
+			b := true
+			filter.Resolved = &b
+		case "0", "false", "no":
+			b := false
+			filter.Resolved = &b
+		default:
+			response.BadRequest(c, "Invalid resolved")
+			return
+		}
+	}
+	if statusCodesStr := strings.TrimSpace(c.Query("status_codes")); statusCodesStr != "" {
+		parts := strings.Split(statusCodesStr, ",")
+		out := make([]int, 0, len(parts))
+		for _, part := range parts {
+			p := strings.TrimSpace(part)
+			if p == "" {
+				continue
+			}
+			n, err := strconv.Atoi(p)
+			if err != nil || n < 0 {
+				response.BadRequest(c, "Invalid status_codes")
+				return
+			}
+			out = append(out, n)
+		}
+		filter.StatusCodes = out
+	}
+
+	result, err := h.opsService.GetErrorLogs(c.Request.Context(), filter)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Paginated(c, result.Errors, int64(result.Total), result.Page, result.PageSize)
+}
+
+// GetUpstreamError returns upstream error detail.
+// GET /api/v1/admin/ops/upstream-errors/:id
+func (h *OpsHandler) GetUpstreamError(c *gin.Context) {
+	h.GetErrorLogByID(c)
+}
+
+// RetryUpstreamError retries upstream error using the original account_id.
+// POST /api/v1/admin/ops/upstream-errors/:id/retry
+func (h *OpsHandler) RetryUpstreamError(c *gin.Context) {
+	if h.opsService == nil {
+		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
+		return
+	}
+	if err := h.opsService.RequireMonitoringEnabled(c.Request.Context()); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	subject, ok := middleware.GetAuthSubjectFromContext(c)
+	if !ok || subject.UserID <= 0 {
+		response.Error(c, http.StatusUnauthorized, "Unauthorized")
+		return
+	}
+
+	idStr := strings.TrimSpace(c.Param("id"))
+	id, err := strconv.ParseInt(idStr, 10, 64)
+	if err != nil || id <= 0 {
+		response.BadRequest(c, "Invalid error id")
+		return
+	}
+
+	result, err := h.opsService.RetryError(c.Request.Context(), subject.UserID, id, service.OpsRetryModeUpstream, nil)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, result)
+}
+
+// ResolveUpstreamError toggles resolved status.
+// PUT /api/v1/admin/ops/upstream-errors/:id/resolve
+func (h *OpsHandler) ResolveUpstreamError(c *gin.Context) {
+	h.UpdateErrorResolution(c)
+}
+
+// ==================== Existing endpoints ====================
+
 // ListRequestDetails returns a request-level list (success + error) for drill-down.
 // GET /api/v1/admin/ops/requests
 func (h *OpsHandler) ListRequestDetails(c *gin.Context) {

backend/internal/server/routes/admin.go

@@ -113,13 +113,26 @@ func registerOpsRoutes(admin *gin.RouterGroup, h *handler.Handlers) {
 			ws.GET("/qps", h.Admin.Ops.QPSWSHandler)
 		}
 
-		// Error logs (MVP-1)
+		// Error logs (legacy)
 		ops.GET("/errors", h.Admin.Ops.GetErrorLogs)
 		ops.GET("/errors/:id", h.Admin.Ops.GetErrorLogByID)
 		ops.GET("/errors/:id/retries", h.Admin.Ops.ListRetryAttempts)
 		ops.POST("/errors/:id/retry", h.Admin.Ops.RetryErrorRequest)
 		ops.PUT("/errors/:id/resolve", h.Admin.Ops.UpdateErrorResolution)
 
+		// Request errors (client-visible failures)
+		ops.GET("/request-errors", h.Admin.Ops.ListRequestErrors)
+		ops.GET("/request-errors/:id", h.Admin.Ops.GetRequestError)
+		ops.POST("/request-errors/:id/retry-client", h.Admin.Ops.RetryRequestErrorClient)
+		ops.POST("/request-errors/:id/upstream-errors/:idx/retry", h.Admin.Ops.RetryRequestErrorUpstreamEvent)
+		ops.PUT("/request-errors/:id/resolve", h.Admin.Ops.ResolveRequestError)
+
+		// Upstream errors (independent upstream failures)
+		ops.GET("/upstream-errors", h.Admin.Ops.ListUpstreamErrors)
+		ops.GET("/upstream-errors/:id", h.Admin.Ops.GetUpstreamError)
+		ops.POST("/upstream-errors/:id/retry", h.Admin.Ops.RetryUpstreamError)
+		ops.PUT("/upstream-errors/:id/resolve", h.Admin.Ops.ResolveUpstreamError)
+
 		// Request drilldown (success + error)
 		ops.GET("/requests", h.Admin.Ops.ListRequestDetails)