feat: merge dev

2026-01-15 15:14:44 +08:00
parent a458e684bc
commit 90bce60b85
107 changed files with 6705 additions and 1611 deletions
--- a/backend/internal/handler/admin/ops_handler.go
+++ b/backend/internal/handler/admin/ops_handler.go
@@ -19,6 +19,57 @@ type OpsHandler struct {
 	opsService *service.OpsService
 }

+// GetErrorLogByID returns ops error log detail.
+// GET /api/v1/admin/ops/errors/:id
+func (h *OpsHandler) GetErrorLogByID(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
+	}
+
+	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
+	}
+
+	detail, err := h.opsService.GetErrorLogByID(c.Request.Context(), id)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	response.Success(c, detail)
+}
+
+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}
 }
@@ -47,16 +98,26 @@ func (h *OpsHandler) GetErrorLogs(c *gin.Context) {
 		return
 	}

-	filter := &service.OpsErrorLogFilter{
-		Page:     page,
-		PageSize: pageSize,
-	}
+	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"))
+	filter.UserQuery = strings.TrimSpace(c.Query("user_query"))
+
+	// 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
@@ -77,11 +138,19 @@ func (h *OpsHandler) GetErrorLogs(c *gin.Context) {
 		}
 		filter.AccountID = &id
 	}
-	if phase := strings.TrimSpace(c.Query("phase")); phase != "" {
-		filter.Phase = phase
-	}
-	if q := strings.TrimSpace(c.Query("q")); q != "" {
-		filter.Query = q
+
+	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, ",")
@@ -106,13 +175,120 @@ func (h *OpsHandler) GetErrorLogs(c *gin.Context) {
 		response.ErrorFrom(c, err)
 		return
 	}
-
 	response.Paginated(c, result.Errors, int64(result.Total), result.Page, result.PageSize)
 }

-// GetErrorLogByID returns a single error log detail.
-// GET /api/v1/admin/ops/errors/:id
-func (h *OpsHandler) GetErrorLogByID(c *gin.Context) {
+// 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"))
+	filter.UserQuery = strings.TrimSpace(c.Query("user_query"))
+
+	// 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)
+}
+
+// ListRequestErrorUpstreamErrors lists upstream error logs correlated to a request error.
+// GET /api/v1/admin/ops/request-errors/:id/upstream-errors
+func (h *OpsHandler) ListRequestErrorUpstreamErrors(c *gin.Context) {
 	if h.opsService == nil {
 		response.Error(c, http.StatusServiceUnavailable, "Ops service not available")
 		return
@@ -129,15 +305,306 @@ func (h *OpsHandler) GetErrorLogByID(c *gin.Context) {
 		return
 	}

+	// Load request error to get correlation keys.
 	detail, err := h.opsService.GetErrorLogByID(c.Request.Context(), id)
 	if err != nil {
 		response.ErrorFrom(c, err)
 		return
 	}

-	response.Success(c, detail)
+	// Correlate by request_id/client_request_id.
+	requestID := strings.TrimSpace(detail.RequestID)
+	clientRequestID := strings.TrimSpace(detail.ClientRequestID)
+	if requestID == "" && clientRequestID == "" {
+		response.Paginated(c, []*service.OpsErrorLog{}, 0, 1, 10)
+		return
+	}
+
+	page, pageSize := response.ParsePagination(c)
+	if pageSize > 500 {
+		pageSize = 500
+	}
+
+	// Keep correlation window wide enough so linked upstream errors
+	// are discoverable even when UI defaults to 1h elsewhere.
+	startTime, endTime, err := parseOpsTimeRange(c, "30d")
+	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 = "all"
+	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
+	}
+
+	// Prefer exact match on request_id; if missing, fall back to client_request_id.
+	if requestID != "" {
+		filter.RequestID = requestID
+	} else {
+		filter.ClientRequestID = clientRequestID
+	}
+
+	result, err := h.opsService.GetErrorLogs(c.Request.Context(), filter)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+
+	// If client asks for details, expand each upstream error log to include upstream response fields.
+	includeDetail := strings.TrimSpace(c.Query("include_detail"))
+	if includeDetail == "1" || strings.EqualFold(includeDetail, "true") || strings.EqualFold(includeDetail, "yes") {
+		details := make([]*service.OpsErrorLogDetail, 0, len(result.Errors))
+		for _, item := range result.Errors {
+			if item == nil {
+				continue
+			}
+			d, err := h.opsService.GetErrorLogByID(c.Request.Context(), item.ID)
+			if err != nil || d == nil {
+				continue
+			}
+			details = append(details, d)
+		}
+		response.Paginated(c, details, int64(result.Total), result.Page, result.PageSize)
+		return
+	}
+
+	response.Paginated(c, result.Errors, int64(result.Total), result.Page, result.PageSize)
 }

+// 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) {
@@ -242,6 +709,11 @@ func (h *OpsHandler) ListRequestDetails(c *gin.Context) {
 type opsRetryRequest struct {
 	Mode            string `json:"mode"`
 	PinnedAccountID *int64 `json:"pinned_account_id"`
+	Force           bool   `json:"force"`
+}
+
+type opsResolveRequest struct {
+	Resolved bool `json:"resolved"`
 }

 // RetryErrorRequest retries a failed request using stored request_body.
@@ -278,6 +750,16 @@ func (h *OpsHandler) RetryErrorRequest(c *gin.Context) {
 		req.Mode = service.OpsRetryModeClient
 	}

+	// Force flag is currently a UI-level acknowledgement. Server may still enforce safety constraints.
+	_ = req.Force
+
+	// Legacy endpoint safety: only allow retrying the client request here.
+	// Upstream retries must go through the split endpoints.
+	if strings.EqualFold(strings.TrimSpace(req.Mode), service.OpsRetryModeUpstream) {
+		response.BadRequest(c, "upstream retry is not supported on this endpoint")
+		return
+	}
+
 	result, err := h.opsService.RetryError(c.Request.Context(), subject.UserID, id, req.Mode, req.PinnedAccountID)
 	if err != nil {
 		response.ErrorFrom(c, err)
@@ -287,6 +769,81 @@ func (h *OpsHandler) RetryErrorRequest(c *gin.Context) {
 	response.Success(c, result)
 }

+// ListRetryAttempts lists retry attempts for an error log.
+// GET /api/v1/admin/ops/errors/:id/retries
+func (h *OpsHandler) ListRetryAttempts(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
+	}
+
+	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
+	}
+
+	limit := 50
+	if v := strings.TrimSpace(c.Query("limit")); v != "" {
+		n, err := strconv.Atoi(v)
+		if err != nil || n <= 0 {
+			response.BadRequest(c, "Invalid limit")
+			return
+		}
+		limit = n
+	}
+
+	items, err := h.opsService.ListRetryAttemptsByErrorID(c.Request.Context(), id, limit)
+	if err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, items)
+}
+
+// UpdateErrorResolution allows manual resolve/unresolve.
+// PUT /api/v1/admin/ops/errors/:id/resolve
+func (h *OpsHandler) UpdateErrorResolution(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
+	}
+
+	var req opsResolveRequest
+	if err := c.ShouldBindJSON(&req); err != nil {
+		response.BadRequest(c, "Invalid request: "+err.Error())
+		return
+	}
+	uid := subject.UserID
+	if err := h.opsService.UpdateErrorResolution(c.Request.Context(), id, req.Resolved, &uid, nil); err != nil {
+		response.ErrorFrom(c, err)
+		return
+	}
+	response.Success(c, gin.H{"ok": true})
+}
+
 func parseOpsTimeRange(c *gin.Context, defaultRange string) (time.Time, time.Time, error) {
 	startStr := strings.TrimSpace(c.Query("start_time"))
 	endStr := strings.TrimSpace(c.Query("end_time"))
@@ -358,6 +915,10 @@ func parseOpsDuration(v string) (time.Duration, bool) {
 		return 6 * time.Hour, true
 	case "24h":
 		return 24 * time.Hour, true
+	case "7d":
+		return 7 * 24 * time.Hour, true
+	case "30d":
+		return 30 * 24 * time.Hour, true
 	default:
 		return 0, false
 	}