feat(generation): migrate article generation and template-assist to RabbitMQ queues

Replace in-process generation with queue-based async execution via
RabbitMQ. Add generation task runtime for lifecycle management, article
generation payload/queue abstractions, and template-assist queue publisher.
Refactor prompt generation service to support batch generation with
per-item error tracking. Update stream hub to bridge queue events to SSE.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

server/internal/bootstrap/bootstrap.go

@@ -94,7 +94,7 @@ func New(configPath string) (*App, error) {
 	vectorStore := retrieval.NewQdrantStore(cfg.Qdrant, logger)
 	objectStorageClient := objectstorage.New(cfg.ObjectStorage, logger)
 	auditLogs := auditlog.NewAsyncWriter(pool, logger)
-	generationStreams := stream.NewGenerationHub()
+	generationStreams := stream.NewGenerationHub(mqClient)
 	appCache := cache.New(cfg.Cache.Driver, rdb)
 
 	if cfg.Server.Mode == "release" {

server/internal/shared/stream/generation.go

@@ -1,8 +1,14 @@
 package stream
 
 import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
 	"sync"
 	"time"
+
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
 )
 
 type GenerationEvent struct {
@@ -18,6 +24,11 @@ type GenerationEvent struct {
 	UpdatedAt time.Time `json:"updated_at"`
 }
 
+type generationEventEnvelope struct {
+	Source string          `json:"source"`
+	Event  GenerationEvent `json:"event"`
+}
+
 type generationStream struct {
 	snapshot       GenerationEvent
 	subscribers    map[int]chan GenerationEvent
@@ -26,17 +37,42 @@ type generationStream struct {
 }
 
 type GenerationHub struct {
-	mu      sync.Mutex
-	streams map[int64]*generationStream
+	mu           sync.Mutex
+	streams      map[int64]*generationStream
+	rabbitMQ     *rabbitmq.Client
+	instanceID   string
+	consumerName string
 }
 
-func NewGenerationHub() *GenerationHub {
+func NewGenerationHub(rabbitMQClient *rabbitmq.Client) *GenerationHub {
+	instanceID := fmt.Sprintf("%d-%d", os.Getpid(), time.Now().UnixNano())
 	return &GenerationHub{
-		streams: make(map[int64]*generationStream),
+		streams:      make(map[int64]*generationStream),
+		rabbitMQ:     rabbitMQClient,
+		instanceID:   instanceID,
+		consumerName: "generation-stream-" + instanceID,
 	}
 }
 
-func (h *GenerationHub) Start(articleID, taskID int64, title string) {
+func (h *GenerationHub) Run(ctx context.Context) {
+	if h == nil || h.rabbitMQ == nil {
+		return
+	}
+	go h.run(ctx)
+}
+
+func (h *GenerationHub) Seed(event GenerationEvent) {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+
+	state := h.ensureLocked(event.ArticleID)
+	applySnapshotLocked(state, event)
+	if state.completed && len(state.subscribers) == 0 {
+		delete(h.streams, event.ArticleID)
+	}
+}
+
+func (h *GenerationHub) StartEvent(articleID, taskID int64, title string) {
 	h.publish(articleID, func(state *generationStream) GenerationEvent {
 		now := time.Now()
 		state.completed = false
@@ -59,6 +95,10 @@ func (h *GenerationHub) Start(articleID, taskID int64, title string) {
 	})
 }
 
+func (h *GenerationHub) Start(articleID, taskID int64, title string) {
+	h.StartEvent(articleID, taskID, title)
+}
+
 func (h *GenerationHub) AppendDelta(articleID, taskID int64, title, delta string) {
 	h.publish(articleID, func(state *generationStream) GenerationEvent {
 		now := time.Now()
@@ -69,6 +109,7 @@ func (h *GenerationHub) AppendDelta(articleID, taskID int64, title, delta string
 		state.snapshot.Status = "generating"
 		state.snapshot.Content += delta
 		state.snapshot.UpdatedAt = now
+		state.completed = false
 
 		return GenerationEvent{
 			Type:      "delta",
@@ -188,14 +229,83 @@ func (h *GenerationHub) Snapshot(articleID int64) (*GenerationEvent, bool) {
 	return &snapshot, true
 }
 
+func (h *GenerationHub) run(ctx context.Context) {
+	for {
+		deliveries, ch, err := h.rabbitMQ.ConsumeGenerationEvents(h.consumerName)
+		if err != nil {
+			select {
+			case <-ctx.Done():
+				return
+			case <-time.After(2 * time.Second):
+				continue
+			}
+		}
+
+		closed := false
+		for !closed {
+			select {
+			case <-ctx.Done():
+				_ = ch.Close()
+				return
+			case delivery, ok := <-deliveries:
+				if !ok {
+					closed = true
+					continue
+				}
+				h.handleDelivery(delivery.Body)
+				_ = delivery.Ack(false)
+			}
+		}
+		_ = ch.Close()
+
+		select {
+		case <-ctx.Done():
+			return
+		case <-time.After(2 * time.Second):
+		}
+	}
+}
+
+func (h *GenerationHub) handleDelivery(body []byte) {
+	var envelope generationEventEnvelope
+	if err := json.Unmarshal(body, &envelope); err != nil {
+		return
+	}
+	if envelope.Source == h.instanceID {
+		return
+	}
+
+	h.mu.Lock()
+	state := h.streams[envelope.Event.ArticleID]
+	if state == nil {
+		h.mu.Unlock()
+		return
+	}
+
+	applyRemoteEventLocked(state, envelope.Event)
+	subscribers := collectSubscribersLocked(state)
+	shouldDelete := state.completed && len(state.subscribers) == 0
+	if shouldDelete {
+		delete(h.streams, envelope.Event.ArticleID)
+	}
+	h.mu.Unlock()
+
+	for _, ch := range subscribers {
+		select {
+		case ch <- envelope.Event:
+		default:
+		}
+	}
+}
+
 func (h *GenerationHub) publish(articleID int64, mutate func(*generationStream) GenerationEvent) {
 	h.mu.Lock()
 	state := h.ensureLocked(articleID)
 	event := mutate(state)
-
-	subscribers := make([]chan GenerationEvent, 0, len(state.subscribers))
-	for _, ch := range state.subscribers {
-		subscribers = append(subscribers, ch)
+	subscribers := collectSubscribersLocked(state)
+	shouldDelete := state.completed && len(state.subscribers) == 0
+	if shouldDelete {
+		delete(h.streams, articleID)
 	}
 	h.mu.Unlock()
 
@@ -205,6 +315,16 @@ func (h *GenerationHub) publish(articleID int64, mutate func(*generationStream)
 		default:
 		}
 	}
+
+	if h.rabbitMQ != nil {
+		payload, err := json.Marshal(generationEventEnvelope{
+			Source: h.instanceID,
+			Event:  event,
+		})
+		if err == nil {
+			_ = h.rabbitMQ.PublishGenerationEvent(context.Background(), payload)
+		}
+	}
 }
 
 func (h *GenerationHub) ensureLocked(articleID int64) *generationStream {
@@ -219,3 +339,78 @@ func (h *GenerationHub) ensureLocked(articleID int64) *generationStream {
 	h.streams[articleID] = state
 	return state
 }
+
+func collectSubscribersLocked(state *generationStream) []chan GenerationEvent {
+	subscribers := make([]chan GenerationEvent, 0, len(state.subscribers))
+	for _, ch := range state.subscribers {
+		subscribers = append(subscribers, ch)
+	}
+	return subscribers
+}
+
+func applySnapshotLocked(state *generationStream, event GenerationEvent) {
+	state.snapshot = GenerationEvent{
+		Type:      "snapshot",
+		ArticleID: event.ArticleID,
+		TaskID:    event.TaskID,
+		Title:     event.Title,
+		Status:    event.Status,
+		Content:   event.Content,
+		Error:     event.Error,
+		Done:      event.Done,
+		UpdatedAt: event.UpdatedAt,
+	}
+	state.completed = event.Done
+}
+
+func applyRemoteEventLocked(state *generationStream, event GenerationEvent) {
+	switch event.Type {
+	case "delta":
+		state.completed = false
+		state.snapshot.Type = "snapshot"
+		state.snapshot.ArticleID = event.ArticleID
+		state.snapshot.TaskID = event.TaskID
+		state.snapshot.Title = event.Title
+		state.snapshot.Status = event.Status
+		if event.Content != "" {
+			state.snapshot.Content = event.Content
+		} else {
+			state.snapshot.Content += event.Delta
+		}
+		state.snapshot.UpdatedAt = event.UpdatedAt
+	case "completed":
+		applySnapshotLocked(state, GenerationEvent{
+			Type:      "snapshot",
+			ArticleID: event.ArticleID,
+			TaskID:    event.TaskID,
+			Title:     event.Title,
+			Status:    "completed",
+			Content:   event.Content,
+			Done:      true,
+			UpdatedAt: event.UpdatedAt,
+		})
+	case "error":
+		applySnapshotLocked(state, GenerationEvent{
+			Type:      "snapshot",
+			ArticleID: event.ArticleID,
+			TaskID:    event.TaskID,
+			Title:     event.Title,
+			Status:    "failed",
+			Error:     event.Error,
+			Done:      true,
+			UpdatedAt: event.UpdatedAt,
+		})
+	default:
+		applySnapshotLocked(state, GenerationEvent{
+			Type:      "snapshot",
+			ArticleID: event.ArticleID,
+			TaskID:    event.TaskID,
+			Title:     event.Title,
+			Status:    event.Status,
+			Content:   event.Content,
+			Error:     event.Error,
+			Done:      event.Done,
+			UpdatedAt: event.UpdatedAt,
+		})
+	}
+}

server/internal/tenant/app/article_generation_payload.go

@@ -0,0 +1,97 @@
+package app
+
+import "github.com/geo-platform/tenant-api/internal/tenant/repository"
+
+const (
+	generationPayloadTemplateKey            = "__snapshot_template_key"
+	generationPayloadTemplateName           = "__snapshot_template_name"
+	generationPayloadTemplatePromptTemplate = "__snapshot_template_prompt_template"
+	generationPayloadPromptRuleID           = "__snapshot_prompt_rule_id"
+	generationPayloadPromptRuleName         = "__snapshot_prompt_rule_name"
+	generationPayloadPromptRuleContent      = "__snapshot_prompt_rule_content"
+	generationPayloadPromptRuleScene        = "__snapshot_prompt_rule_scene"
+	generationPayloadPromptRuleTone         = "__snapshot_prompt_rule_tone"
+	generationPayloadPromptRuleWordCount    = "__snapshot_prompt_rule_word_count"
+	generationPayloadKnowledgePrompt        = "__snapshot_knowledge_prompt"
+)
+
+func attachTemplateSnapshot(payload map[string]interface{}, record *repository.TemplateRecord) map[string]interface{} {
+	if payload == nil {
+		payload = map[string]interface{}{}
+	}
+	if record == nil {
+		return payload
+	}
+
+	payload[generationPayloadTemplateKey] = record.TemplateKey
+	payload[generationPayloadTemplateName] = record.TemplateName
+	if record.PromptTemplate != nil {
+		payload[generationPayloadTemplatePromptTemplate] = *record.PromptTemplate
+	}
+
+	return payload
+}
+
+func extractTemplateSnapshot(payload map[string]interface{}) (string, string, *string) {
+	templateKey := extractString(payload, generationPayloadTemplateKey)
+	templateName := extractString(payload, generationPayloadTemplateName)
+	promptTemplateText := extractString(payload, generationPayloadTemplatePromptTemplate)
+	if templateKey == "" || templateName == "" {
+		return "", "", nil
+	}
+	if promptTemplateText == "" {
+		return templateKey, templateName, nil
+	}
+	return templateKey, templateName, &promptTemplateText
+}
+
+func attachPromptRuleSnapshot(payload map[string]interface{}, rule *promptRuleGenerationRecord) map[string]interface{} {
+	if payload == nil {
+		payload = map[string]interface{}{}
+	}
+	if rule == nil {
+		return payload
+	}
+
+	payload[generationPayloadPromptRuleID] = rule.ID
+	payload[generationPayloadPromptRuleName] = rule.Name
+	payload[generationPayloadPromptRuleContent] = rule.PromptContent
+	if rule.Scene != nil {
+		payload[generationPayloadPromptRuleScene] = *rule.Scene
+	}
+	if rule.DefaultTone != nil {
+		payload[generationPayloadPromptRuleTone] = *rule.DefaultTone
+	}
+	if rule.DefaultWordCount != nil {
+		payload[generationPayloadPromptRuleWordCount] = *rule.DefaultWordCount
+	}
+
+	return payload
+}
+
+func extractPromptRuleSnapshot(payload map[string]interface{}) *promptRuleGenerationRecord {
+	ruleID := int64(extractInt(payload, generationPayloadPromptRuleID))
+	name := extractString(payload, generationPayloadPromptRuleName)
+	content := extractString(payload, generationPayloadPromptRuleContent)
+	if ruleID <= 0 || content == "" {
+		return nil
+	}
+
+	record := &promptRuleGenerationRecord{
+		ID:            ruleID,
+		Name:          name,
+		PromptContent: content,
+		Status:        "enabled",
+	}
+	if scene := extractString(payload, generationPayloadPromptRuleScene); scene != "" {
+		record.Scene = &scene
+	}
+	if tone := extractString(payload, generationPayloadPromptRuleTone); tone != "" {
+		record.DefaultTone = &tone
+	}
+	if wordCount := extractInt(payload, generationPayloadPromptRuleWordCount); wordCount > 0 {
+		value := wordCount
+		record.DefaultWordCount = &value
+	}
+	return record
+}

server/internal/tenant/app/article_generation_queue.go

@@ -0,0 +1,63 @@
+package app
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
+)
+
+type articleGenerationTaskEnvelope struct {
+	TaskID    int64  `json:"task_id"`
+	TaskType  string `json:"task_type"`
+	TenantID  int64  `json:"tenant_id"`
+	ArticleID int64  `json:"article_id"`
+}
+
+type ArticleGenerationTaskEnvelope struct {
+	TaskID    int64  `json:"task_id"`
+	TaskType  string `json:"task_type"`
+	TenantID  int64  `json:"tenant_id"`
+	ArticleID int64  `json:"article_id"`
+}
+
+func publishArticleGenerationTask(ctx context.Context, client *rabbitmq.Client, envelope articleGenerationTaskEnvelope) error {
+	if client == nil {
+		return fmt.Errorf("rabbitmq client is not configured")
+	}
+
+	payload, err := json.Marshal(envelope)
+	if err != nil {
+		return fmt.Errorf("marshal generation task envelope: %w", err)
+	}
+
+	if err := client.PublishGenerationTask(ctx, payload); err != nil {
+		return fmt.Errorf("publish generation task: %w", err)
+	}
+	return nil
+}
+
+func decodeArticleGenerationTaskEnvelope(payload []byte) (*articleGenerationTaskEnvelope, error) {
+	var envelope articleGenerationTaskEnvelope
+	if err := json.Unmarshal(payload, &envelope); err != nil {
+		return nil, fmt.Errorf("decode generation task envelope: %w", err)
+	}
+	if envelope.TaskID <= 0 {
+		return nil, fmt.Errorf("generation task envelope task_id is required")
+	}
+	return &envelope, nil
+}
+
+func DecodeArticleGenerationTaskEnvelope(payload []byte) (*ArticleGenerationTaskEnvelope, error) {
+	envelope, err := decodeArticleGenerationTaskEnvelope(payload)
+	if err != nil {
+		return nil, err
+	}
+	return &ArticleGenerationTaskEnvelope{
+		TaskID:    envelope.TaskID,
+		TaskType:  envelope.TaskType,
+		TenantID:  envelope.TenantID,
+		ArticleID: envelope.ArticleID,
+	}, nil
+}

server/internal/tenant/app/generation_task_runtime.go

@@ -0,0 +1,222 @@
+package app
+
+import (
+	"context"
+	"fmt"
+	"strings"
+	"time"
+
+	"github.com/jackc/pgx/v5/pgxpool"
+
+	"github.com/geo-platform/tenant-api/internal/shared/llm"
+	"github.com/geo-platform/tenant-api/internal/tenant/repository"
+)
+
+type ClaimedGenerationTask struct {
+	ID                 int64
+	TenantID           int64
+	OperatorID         int64
+	ArticleID          int64
+	QuotaReservationID int64
+	TaskType           string
+	InputParams        map[string]interface{}
+}
+
+func HandleClaimedGenerationTaskFailure(
+	ctx context.Context,
+	pool *pgxpool.Pool,
+	templateService *TemplateService,
+	promptRuleService *PromptRuleGenerationService,
+	task ClaimedGenerationTask,
+	taskErr error,
+) error {
+	errMsg := formatGenerationFailure("task_load", taskErr)
+	now := time.Now()
+
+	auditRepo := repository.NewAuditRepository(pool)
+	articleRepo := repository.NewArticleRepository(pool)
+	quotaRepo := repository.NewQuotaRepository(pool)
+
+	_ = auditRepo.UpdateGenerationTaskStatus(ctx, repository.GenerationTaskStatusInput{
+		ID:           task.ID,
+		TenantID:     task.TenantID,
+		Status:       "failed",
+		ErrorMessage: &errMsg,
+		CompletedAt:  &now,
+	})
+	if task.ArticleID > 0 {
+		_ = articleRepo.UpdateArticleGenerateStatus(ctx, task.ArticleID, task.TenantID, "failed")
+	}
+
+	if task.QuotaReservationID > 0 {
+		balance, balanceErr := quotaRepo.GetCurrentBalance(ctx, task.TenantID, "article_generation")
+		if balanceErr == nil && task.OperatorID > 0 {
+			reason := "generation_refund"
+			referenceType := "generation_task"
+			taskReferenceID := task.ID
+			_, _ = quotaRepo.InsertQuotaLedger(ctx, repository.QuotaLedgerInput{
+				TenantID:      task.TenantID,
+				OperatorID:    task.OperatorID,
+				QuotaType:     "article_generation",
+				Delta:         1,
+				BalanceAfter:  balance + 1,
+				Reason:        &reason,
+				ReferenceType: &referenceType,
+				ReferenceID:   &taskReferenceID,
+			})
+		}
+
+		_ = quotaRepo.RefundReservation(ctx, task.QuotaReservationID, task.TenantID)
+	}
+
+	title := resolveArticleTitle(task.InputParams, "")
+	if strings.TrimSpace(extractString(task.InputParams, "task_name")) != "" {
+		title = strings.TrimSpace(extractString(task.InputParams, "task_name"))
+	}
+	if title == "" {
+		title = promptRuleGeneratingTitlePlaceholder
+	}
+
+	if task.TaskType == "template" && templateService != nil && templateService.streamEnabled {
+		templateService.streams.Fail(task.ArticleID, task.ID, title, errMsg)
+	}
+	if task.TaskType == "custom_generation" && promptRuleService != nil && promptRuleService.streamEnabled {
+		promptRuleService.streams.Fail(task.ArticleID, task.ID, title, errMsg)
+	}
+
+	return nil
+}
+
+func (s *TemplateService) ProcessQueuedTask(ctx context.Context, task ClaimedGenerationTask) {
+	job := generationJob{
+		OperatorID:    task.OperatorID,
+		TenantID:      task.TenantID,
+		ArticleID:     task.ArticleID,
+		TaskID:        task.ID,
+		ReservationID: task.QuotaReservationID,
+		Params:        cloneInputParams(task.InputParams),
+		InitialTitle:  resolveArticleTitle(task.InputParams, ""),
+		WebSearch: llm.WebSearchOptions{
+			Enabled: extractBool(task.InputParams, "enable_web_search"),
+		},
+	}
+	if limit := extractInt(task.InputParams, "web_search_limit"); limit > 0 {
+		job.WebSearch.Limit = int32(limit)
+	}
+
+	templateKey, templateName, promptTemplate := extractTemplateSnapshot(task.InputParams)
+	if templateKey == "" || templateName == "" {
+		record, err := s.loadTemplateRecordForTask(ctx, task)
+		if err != nil {
+			s.failGeneration(context.Background(), job, job.InitialTitle, "task_load", err)
+			return
+		}
+		templateKey = record.TemplateKey
+		templateName = record.TemplateName
+		promptTemplate = record.PromptTemplate
+	}
+	job.TemplateKey = templateKey
+	job.TemplateName = templateName
+	job.PromptTemplate = promptTemplate
+
+	s.executeGeneration(ctx, job)
+}
+
+func (s *TemplateService) loadTemplateRecordForTask(ctx context.Context, task ClaimedGenerationTask) (*repository.TemplateRecord, error) {
+	var templateID int64
+	if err := s.pool.QueryRow(ctx, `
+		SELECT template_id
+		FROM articles
+		WHERE id = $1
+		  AND tenant_id = $2
+		  AND source_type = 'template'
+		  AND deleted_at IS NULL
+	`, task.ArticleID, task.TenantID).Scan(&templateID); err != nil {
+		return nil, fmt.Errorf("load task template id: %w", err)
+	}
+
+	record, err := s.templates.GetTemplateByID(ctx, templateID, task.TenantID)
+	if err != nil {
+		return nil, fmt.Errorf("load task template: %w", err)
+	}
+	return record, nil
+}
+
+func (s *PromptRuleGenerationService) ProcessQueuedTask(ctx context.Context, task ClaimedGenerationTask) {
+	inputParams := cloneInputParams(task.InputParams)
+	initialTitle := strings.TrimSpace(extractString(inputParams, "task_name"))
+	if initialTitle == "" {
+		initialTitle = promptRuleGeneratingTitlePlaceholder
+	}
+
+	job := promptRuleGenerationJob{
+		OperatorID:    task.OperatorID,
+		TenantID:      task.TenantID,
+		ArticleID:     task.ArticleID,
+		TaskID:        task.ID,
+		ReservationID: task.QuotaReservationID,
+		InputParams:   inputParams,
+		InitialTitle:  initialTitle,
+		WebSearch: llm.WebSearchOptions{
+			Enabled: extractBool(inputParams, "enable_web_search"),
+		},
+	}
+
+	rule := extractPromptRuleSnapshot(inputParams)
+	if rule == nil {
+		promptRuleID, err := s.resolvePromptRuleIDForTask(ctx, task)
+		if err != nil {
+			s.failGeneration(context.Background(), job, initialTitle, "task_load", err)
+			return
+		}
+		rule, err = s.loadPromptRule(ctx, task.TenantID, promptRuleID)
+		if err != nil {
+			s.failGeneration(context.Background(), job, initialTitle, "task_load", err)
+			return
+		}
+	}
+
+	knowledgePrompt := strings.TrimSpace(extractString(inputParams, generationPayloadKnowledgePrompt))
+	if knowledgePrompt == "" {
+		if knowledgeGroupIDs := extractKnowledgeGroupIDs(inputParams["knowledge_group_ids"]); len(knowledgeGroupIDs) > 0 {
+			if s.knowledge == nil {
+				s.failGeneration(context.Background(), job, initialTitle, "knowledge_resolve", fmt.Errorf("knowledge service is not configured"))
+				return
+			}
+
+			knowledgeContext, err := s.knowledge.ResolveContext(
+				ctx,
+				task.TenantID,
+				knowledgeGroupIDs,
+				buildPromptRuleKnowledgeQuery(rule, inputParams),
+			)
+			if err != nil {
+				s.failGeneration(context.Background(), job, initialTitle, "knowledge_resolve", err)
+				return
+			}
+			knowledgePrompt = s.knowledge.RenderPromptSection(knowledgeContext)
+		}
+	}
+
+	job.Prompt = buildPromptRuleGenerationPrompt(rule, inputParams, knowledgePrompt)
+	s.executeGeneration(ctx, job)
+}
+
+func (s *PromptRuleGenerationService) resolvePromptRuleIDForTask(ctx context.Context, task ClaimedGenerationTask) (int64, error) {
+	if promptRuleID := int64(extractInt(task.InputParams, "prompt_rule_id")); promptRuleID > 0 {
+		return promptRuleID, nil
+	}
+
+	var promptRuleID int64
+	if err := s.pool.QueryRow(ctx, `
+		SELECT prompt_rule_id
+		FROM articles
+		WHERE id = $1
+		  AND tenant_id = $2
+		  AND source_type = 'custom_generation'
+		  AND deleted_at IS NULL
+	`, task.ArticleID, task.TenantID).Scan(&promptRuleID); err != nil {
+		return 0, fmt.Errorf("load task prompt rule id: %w", err)
+	}
+	return promptRuleID, nil
+}

server/internal/tenant/app/knowledge_service.go

@@ -219,7 +219,6 @@ func NewKnowledgeService(
 			go svc.runParseWorker()
 		}
 		go svc.recoverPendingParseTasks()
-		go svc.runDeletedCleanupWorker()
 	}
 
 	return svc
@@ -1753,17 +1752,6 @@ func (s *KnowledgeService) cleanupDeletedKnowledgeItemAsync(tenantID, itemID int
 	}()
 }
 
-func (s *KnowledgeService) runDeletedCleanupWorker() {
-	s.scheduleDeletedKnowledgeCleanup()
-
-	ticker := time.NewTicker(defaultKnowledgeCleanupPoll)
-	defer ticker.Stop()
-
-	for range ticker.C {
-		s.scheduleDeletedKnowledgeCleanup()
-	}
-}
-
 func (s *KnowledgeService) scheduleDeletedKnowledgeCleanup() {
 	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
 	defer cancel()
@@ -1778,6 +1766,13 @@ func (s *KnowledgeService) scheduleDeletedKnowledgeCleanup() {
 	}
 }
 
+func (s *KnowledgeService) RunDeletedCleanupSweep() {
+	if s == nil {
+		return
+	}
+	s.scheduleDeletedKnowledgeCleanup()
+}
+
 func (s *KnowledgeService) listKnowledgeCleanupCandidates(
 	ctx context.Context,
 	limit int,

server/internal/tenant/app/prompt_generate_service.go

@@ -13,6 +13,7 @@ import (
 	"github.com/geo-platform/tenant-api/internal/shared/auth"
 	"github.com/geo-platform/tenant-api/internal/shared/config"
 	"github.com/geo-platform/tenant-api/internal/shared/llm"
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
 	"github.com/geo-platform/tenant-api/internal/shared/response"
 	"github.com/geo-platform/tenant-api/internal/shared/stream"
 	"github.com/geo-platform/tenant-api/internal/tenant/prompts"
@@ -22,12 +23,12 @@ import (
 type PromptRuleGenerationService struct {
 	pool            *pgxpool.Pool
 	llm             llm.Client
+	rabbitMQ        *rabbitmq.Client
 	knowledge       *KnowledgeService
 	streams         *stream.GenerationHub
 	streamEnabled   bool
 	articleTimeout  time.Duration
 	maxOutputTokens int64
-	jobs            chan promptRuleGenerationJob
 }
 
 type promptRuleGenerationJob struct {
@@ -50,10 +51,34 @@ type GenerateFromRuleRequest struct {
 }
 
 type GenerateFromRuleResponse struct {
+	ArticleID              int64                  `json:"article_id"`
+	TaskID                 int64                  `json:"task_id"`
+	ArticleIDs             []int64                `json:"article_ids"`
+	TaskIDs                []int64                `json:"task_ids"`
+	Items                  []GenerateFromRuleItem `json:"items"`
+	RequestedGenerateCount int                    `json:"requested_generate_count"`
+	SuccessCount           int                    `json:"success_count"`
+	FailedCount            int                    `json:"failed_count"`
+}
+
+type GenerateFromRuleItem struct {
 	ArticleID int64 `json:"article_id"`
 	TaskID    int64 `json:"task_id"`
 }
 
+type SchedulePromptRuleGenerationInput struct {
+	ScheduleTaskID  int64
+	OperatorID      *int64
+	TenantID        int64
+	PromptRuleID    int64
+	BrandID         *int64
+	Name            string
+	TargetPlatform  *string
+	EnableWebSearch bool
+	GenerateCount   int
+	ScheduledFor    time.Time
+}
+
 type promptRuleGenerationRecord struct {
 	ID                int64
 	Name              string
@@ -68,24 +93,25 @@ type promptRuleGenerationRecord struct {
 
 const promptRuleGeneratingTitlePlaceholder = "标题生成中"
 
+type enqueuePromptRuleGenerationInput struct {
+	OperatorID     *int64
+	TenantID       int64
+	PromptRuleID   int64
+	BrandID        *int64
+	TargetPlatform *string
+	ExtraParams    map[string]interface{}
+	FallbackName   string
+}
+
 func NewPromptRuleGenerationService(
 	pool *pgxpool.Pool,
 	llmClient llm.Client,
+	rabbitMQClient *rabbitmq.Client,
 	knowledge *KnowledgeService,
 	streams *stream.GenerationHub,
 	cfg config.GenerationConfig,
 	llmMaxOutputTokens int64,
 ) *PromptRuleGenerationService {
-	queueSize := cfg.QueueSize
-	if queueSize <= 0 {
-		queueSize = 128
-	}
-
-	workerCount := cfg.WorkerConcurrency
-	if workerCount <= 0 {
-		workerCount = 1
-	}
-
 	articleTimeout := cfg.ArticleTimeout
 	if articleTimeout <= 0 {
 		articleTimeout = 8 * time.Minute
@@ -94,16 +120,12 @@ func NewPromptRuleGenerationService(
 	svc := &PromptRuleGenerationService{
 		pool:            pool,
 		llm:             llmClient,
+		rabbitMQ:        rabbitMQClient,
 		knowledge:       knowledge,
 		streams:         streams,
 		streamEnabled:   cfg.StreamEnabled,
 		articleTimeout:  articleTimeout,
 		maxOutputTokens: llmMaxOutputTokens,
-		jobs:            make(chan promptRuleGenerationJob, queueSize),
-	}
-
-	for i := 0; i < workerCount; i++ {
-		go svc.runGenerationWorker()
 	}
 
 	return svc
@@ -111,12 +133,86 @@ func NewPromptRuleGenerationService(
 
 func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req GenerateFromRuleRequest) (*GenerateFromRuleResponse, error) {
 	actor := auth.MustActor(ctx)
+	extra := clonePromptRuleExtraParams(req.ExtraParams)
+	if strings.TrimSpace(extractString(extra, "generation_mode")) == "" {
+		extra["generation_mode"] = "instant"
+	}
+	generateCount, err := normalizePromptRuleGenerateCount(extractGenerateCount(extra))
+	if err != nil {
+		return nil, err
+	}
 
+	quotaRepo := repository.NewQuotaRepository(s.pool)
+	balance, err := quotaRepo.GetCurrentBalance(ctx, actor.TenantID, "article_generation")
+	if err != nil || balance < generateCount {
+		return nil, response.ErrForbidden(40301, "quota_insufficient", "generation quota insufficient, please upgrade your plan")
+	}
+
+	results := make([]GenerateFromRuleItem, 0, generateCount)
+	var lastErr error
+	for idx := 0; idx < generateCount; idx++ {
+		result, enqueueErr := s.enqueuePromptRuleGeneration(ctx, enqueuePromptRuleGenerationInput{
+			OperatorID:     &actor.UserID,
+			TenantID:       actor.TenantID,
+			PromptRuleID:   req.PromptRuleID,
+			BrandID:        req.BrandID,
+			TargetPlatform: req.TargetPlatform,
+			ExtraParams:    clonePromptRuleExtraParams(extra),
+			FallbackName:   "即时任务",
+		})
+		if enqueueErr != nil {
+			lastErr = enqueueErr
+			continue
+		}
+		results = append(results, GenerateFromRuleItem{
+			ArticleID: result.ArticleID,
+			TaskID:    result.TaskID,
+		})
+	}
+
+	if len(results) == 0 {
+		if lastErr != nil {
+			return nil, lastErr
+		}
+		return nil, response.ErrInternal(50012, "create_failed", "failed to create generation task")
+	}
+
+	return buildGenerateFromRuleResponse(results, generateCount), nil
+}
+
+func (s *PromptRuleGenerationService) EnqueueScheduledGeneration(ctx context.Context, input SchedulePromptRuleGenerationInput) (*GenerateFromRuleResponse, error) {
+	extra := map[string]interface{}{
+		"generation_mode":   "schedule",
+		"schedule_task_id":  input.ScheduleTaskID,
+		"enable_web_search": input.EnableWebSearch,
+	}
+	if input.GenerateCount > 0 {
+		extra["generate_count"] = input.GenerateCount
+	}
+	if name := strings.TrimSpace(input.Name); name != "" {
+		extra["task_name"] = name
+	}
+	if !input.ScheduledFor.IsZero() {
+		extra["scheduled_for"] = input.ScheduledFor.UTC().Format(time.RFC3339)
+	}
+
+	return s.enqueuePromptRuleGeneration(ctx, enqueuePromptRuleGenerationInput{
+		OperatorID:     input.OperatorID,
+		TenantID:       input.TenantID,
+		PromptRuleID:   input.PromptRuleID,
+		BrandID:        input.BrandID,
+		TargetPlatform: input.TargetPlatform,
+		ExtraParams:    extra,
+		FallbackName:   "定时任务",
+	})
+}
+
+func (s *PromptRuleGenerationService) enqueuePromptRuleGeneration(ctx context.Context, input enqueuePromptRuleGenerationInput) (*GenerateFromRuleResponse, error) {
 	if err := s.llm.Validate(); err != nil {
 		return nil, response.ErrServiceUnavailable(50311, "llm_unavailable", err.Error())
 	}
 
-	rule, err := s.loadPromptRule(ctx, actor.TenantID, req.PromptRuleID)
+	rule, err := s.loadPromptRule(ctx, input.TenantID, input.PromptRuleID)
 	if err != nil {
 		return nil, err
 	}
@@ -124,9 +220,11 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 		return nil, response.ErrBadRequest(40017, "prompt_rule_disabled", "prompt rule is disabled")
 	}
 
-	extra := req.ExtraParams
-	if extractGenerateCount(extra) > 1 {
-		return nil, response.ErrBadRequest(40018, "generate_count_not_supported", "prompt rule instant generation currently supports generate_count = 1 only")
+	extra := input.ExtraParams
+	generationMode := strings.TrimSpace(extractString(extra, "generation_mode"))
+	generateCount, err := normalizePromptRuleGenerateCount(extractGenerateCount(extra))
+	if err != nil {
+		return nil, err
 	}
 
 	taskName := strings.TrimSpace(extractString(extra, "task_name"))
@@ -134,34 +232,46 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 		taskName = strings.TrimSpace(rule.Name)
 	}
 	if taskName == "" {
-		taskName = "即时任务"
+		taskName = strings.TrimSpace(input.FallbackName)
+	}
+	if taskName == "" {
+		taskName = "内容任务"
 	}
 
 	quotaRepo := repository.NewQuotaRepository(s.pool)
-	balance, err := quotaRepo.GetCurrentBalance(ctx, actor.TenantID, "article_generation")
+	balance, err := quotaRepo.GetCurrentBalance(ctx, input.TenantID, "article_generation")
 	if err != nil || balance < 1 {
 		return nil, response.ErrForbidden(40301, "quota_insufficient", "generation quota insufficient, please upgrade your plan")
 	}
 
 	inputParams := map[string]interface{}{
-		"prompt_rule_id":    req.PromptRuleID,
+		"prompt_rule_id":    input.PromptRuleID,
 		"task_name":         taskName,
-		"target_platform":   strings.TrimSpace(derefString(req.TargetPlatform)),
+		"target_platform":   strings.TrimSpace(derefString(input.TargetPlatform)),
 		"enable_web_search": extractBool(extra, "enable_web_search"),
+		"generate_count":    generateCount,
 	}
-	if generationMode := strings.TrimSpace(extractString(extra, "generation_mode")); generationMode != "" {
+	if generationMode != "" {
 		inputParams["generation_mode"] = generationMode
 	}
-	platformIDs := parsePlatformIDs(derefString(req.TargetPlatform))
+	if scheduleTaskID := extractInt(extra, "schedule_task_id"); scheduleTaskID > 0 {
+		inputParams["schedule_task_id"] = scheduleTaskID
+	}
+	if scheduledFor := strings.TrimSpace(extractString(extra, "scheduled_for")); scheduledFor != "" {
+		inputParams["scheduled_for"] = scheduledFor
+	}
+
+	platformIDs := parsePlatformIDs(derefString(input.TargetPlatform))
 	if len(platformIDs) > 0 {
 		inputParams["target_platforms"] = platformIDs
 	}
-	if req.BrandID != nil {
-		inputParams["brand_id"] = *req.BrandID
+	if input.BrandID != nil {
+		inputParams["brand_id"] = *input.BrandID
 	}
 	if len(rule.KnowledgeGroupIDs) > 0 {
 		inputParams["knowledge_group_ids"] = rule.KnowledgeGroupIDs
 	}
+	inputParams = attachPromptRuleSnapshot(inputParams, rule)
 
 	knowledgePrompt := ""
 	if len(rule.KnowledgeGroupIDs) > 0 {
@@ -170,7 +280,7 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 		}
 		knowledgeContext, resolveErr := s.knowledge.ResolveContext(
 			ctx,
-			actor.TenantID,
+			input.TenantID,
 			rule.KnowledgeGroupIDs,
 			buildPromptRuleKnowledgeQuery(rule, inputParams),
 		)
@@ -178,6 +288,9 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 			return nil, resolveErr
 		}
 		knowledgePrompt = s.knowledge.RenderPromptSection(knowledgeContext)
+		if knowledgePrompt != "" {
+			inputParams[generationPayloadKnowledgePrompt] = knowledgePrompt
+		}
 	}
 
 	inputJSON, _ := json.Marshal(inputParams)
@@ -193,13 +306,17 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 
 	quotaTx := repository.NewQuotaRepository(tx)
 	auditTx := repository.NewAuditRepository(tx)
+	operatorID := int64(0)
+	if input.OperatorID != nil {
+		operatorID = *input.OperatorID
+	}
 
 	newBalance := balance - 1
 	reason := "generation_reserve"
 	referenceType := "generation_task"
 	_, err = quotaTx.InsertQuotaLedger(ctx, repository.QuotaLedgerInput{
-		TenantID:      actor.TenantID,
-		OperatorID:    actor.UserID,
+		TenantID:      input.TenantID,
+		OperatorID:    operatorID,
 		QuotaType:     "article_generation",
 		Delta:         -1,
 		BalanceAfter:  newBalance,
@@ -211,8 +328,8 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 	}
 
 	reservationID, err := quotaTx.CreateQuotaReservation(ctx, repository.QuotaReservationInput{
-		TenantID:       actor.TenantID,
-		OperatorID:     actor.UserID,
+		TenantID:       input.TenantID,
+		OperatorID:     operatorID,
 		QuotaType:      "article_generation",
 		ResourceType:   "article",
 		ResourceID:     nil,
@@ -228,18 +345,18 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 		INSERT INTO articles (tenant_id, source_type, prompt_rule_id, generate_status, publish_status, wizard_state_json)
 		VALUES ($1, 'custom_generation', $2, 'generating', 'unpublished', $3)
 		RETURNING id
-	`, actor.TenantID, req.PromptRuleID, wizardStateJSON).Scan(&articleID)
+	`, input.TenantID, input.PromptRuleID, wizardStateJSON).Scan(&articleID)
 	if err != nil {
 		return nil, response.ErrInternal(50012, "create_failed", "failed to create prompt rule article")
 	}
 
-	if err := quotaTx.UpdateQuotaReservationResource(ctx, reservationID, actor.TenantID, articleID); err != nil {
+	if err := quotaTx.UpdateQuotaReservationResource(ctx, reservationID, input.TenantID, articleID); err != nil {
 		return nil, response.ErrInternal(50012, "create_failed", "failed to update quota reservation resource")
 	}
 
 	taskID, err := auditTx.CreateGenerationTask(ctx, repository.GenerationTaskInput{
-		OperatorID:         actor.UserID,
-		TenantID:           actor.TenantID,
+		OperatorID:         operatorID,
+		TenantID:           input.TenantID,
 		ArticleID:          &articleID,
 		QuotaReservationID: &reservationID,
 		TaskType:           "custom_generation",
@@ -254,25 +371,26 @@ func (s *PromptRuleGenerationService) GenerateFromRule(ctx context.Context, req
 	}
 
 	job := promptRuleGenerationJob{
-		OperatorID:    actor.UserID,
-		TenantID:      actor.TenantID,
+		OperatorID:    operatorID,
+		TenantID:      input.TenantID,
 		ArticleID:     articleID,
 		TaskID:        taskID,
 		ReservationID: reservationID,
-		Prompt:        buildPromptRuleGenerationPrompt(rule, inputParams, knowledgePrompt),
 		InputParams:   inputParams,
 		InitialTitle:  promptRuleGeneratingTitlePlaceholder,
-		WebSearch: llm.WebSearchOptions{
-			Enabled: extractBool(extra, "enable_web_search"),
-		},
 	}
 
-	if err := s.enqueueGeneration(job); err != nil {
-		s.failGeneration(context.Background(), job, job.InitialTitle, "queue_enqueue", err)
+	if err := publishArticleGenerationTask(ctx, s.rabbitMQ, articleGenerationTaskEnvelope{
+		TaskID:    taskID,
+		TaskType:  "custom_generation",
+		TenantID:  input.TenantID,
+		ArticleID: articleID,
+	}); err != nil {
+		s.failGeneration(context.Background(), job, promptRuleGeneratingTitlePlaceholder, "queue_enqueue", err)
 		return nil, response.ErrServiceUnavailable(50310, "generation_queue_unavailable", "generation queue is busy, please retry")
 	}
 
-	if s.streamEnabled {
+	if s.streamEnabled && s.streams != nil {
 		s.streams.Start(articleID, taskID, promptRuleGeneratingTitlePlaceholder)
 	}
 
@@ -322,21 +440,6 @@ func (s *PromptRuleGenerationService) loadPromptRule(ctx context.Context, tenant
 	return record, nil
 }
 
-func (s *PromptRuleGenerationService) enqueueGeneration(job promptRuleGenerationJob) error {
-	select {
-	case s.jobs <- job:
-		return nil
-	default:
-		return fmt.Errorf("generation queue is full")
-	}
-}
-
-func (s *PromptRuleGenerationService) runGenerationWorker() {
-	for job := range s.jobs {
-		s.executeGeneration(context.Background(), job)
-	}
-}
-
 func (s *PromptRuleGenerationService) executeGeneration(ctx context.Context, job promptRuleGenerationJob) {
 	now := time.Now()
 	title := strings.TrimSpace(job.InitialTitle)
@@ -557,6 +660,50 @@ func extractGenerateCount(extra map[string]interface{}) int {
 	return value
 }
 
+func normalizePromptRuleGenerateCount(value int) (int, error) {
+	if value < 1 || value > 20 {
+		return 0, response.ErrBadRequest(40022, "invalid_generate_count", "generate_count must be between 1 and 20")
+	}
+	return value, nil
+}
+
+func clonePromptRuleExtraParams(source map[string]interface{}) map[string]interface{} {
+	if len(source) == 0 {
+		return map[string]interface{}{}
+	}
+
+	cloned := make(map[string]interface{}, len(source))
+	for key, value := range source {
+		cloned[key] = value
+	}
+	return cloned
+}
+
+func buildGenerateFromRuleResponse(items []GenerateFromRuleItem, requestedCount int) *GenerateFromRuleResponse {
+	responseItems := make([]GenerateFromRuleItem, 0, len(items))
+	articleIDs := make([]int64, 0, len(items))
+	taskIDs := make([]int64, 0, len(items))
+	for _, item := range items {
+		responseItems = append(responseItems, item)
+		articleIDs = append(articleIDs, item.ArticleID)
+		taskIDs = append(taskIDs, item.TaskID)
+	}
+
+	result := &GenerateFromRuleResponse{
+		ArticleIDs:             articleIDs,
+		TaskIDs:                taskIDs,
+		Items:                  responseItems,
+		RequestedGenerateCount: requestedCount,
+		SuccessCount:           len(items),
+		FailedCount:            requestedCount - len(items),
+	}
+	if len(items) > 0 {
+		result.ArticleID = items[0].ArticleID
+		result.TaskID = items[0].TaskID
+	}
+	return result
+}
+
 func extractBool(extra map[string]interface{}, key string) bool {
 	if len(extra) == 0 {
 		return false

server/internal/tenant/app/template_assist.go

@@ -120,6 +120,8 @@ type assistJob struct {
 	OutlineRequest *OutlineTaskRequest
 }
 
+type AssistJob = assistJob
+
 func (s *TemplateService) CreateAnalyzeTask(ctx context.Context, templateID int64, req AnalyzeTaskRequest) (*AssistTaskCreateResponse, error) {
 	actor := auth.MustActor(ctx)
 
@@ -164,7 +166,7 @@ func (s *TemplateService) CreateAnalyzeTask(ctx context.Context, templateID int6
 		AnalyzePrompt:  extractAnalyzePromptTemplate(templateRecord.CardConfigJSON),
 		AnalyzeRequest: &req,
 	}
-	if err := s.enqueueAssist(job); err != nil {
+	if err := publishTemplateAssistTask(ctx, s.rabbitMQ, job); err != nil {
 		_ = repo.FailTask(context.Background(), taskID, templateAssistTaskTypeAnalyze, actor.TenantID, err.Error(), time.Now())
 		return nil, response.ErrServiceUnavailable(50312, "analyze_queue_unavailable", "analyze queue is busy, please retry")
 	}
@@ -245,7 +247,7 @@ func (s *TemplateService) CreateTitleTask(ctx context.Context, templateID int64,
 		TitlePrompt:  extractTitlePromptTemplate(templateRecord.CardConfigJSON),
 		TitleRequest: &req,
 	}
-	if err := s.enqueueAssist(job); err != nil {
+	if err := publishTemplateAssistTask(ctx, s.rabbitMQ, job); err != nil {
 		_ = repo.FailTask(context.Background(), taskID, templateAssistTaskTypeTitle, actor.TenantID, err.Error(), time.Now())
 		return nil, response.ErrServiceUnavailable(50313, "title_queue_unavailable", "title queue is busy, please retry")
 	}
@@ -326,7 +328,7 @@ func (s *TemplateService) CreateOutlineTask(ctx context.Context, templateID int6
 		OutlinePrompt:  extractOutlinePromptTemplate(templateRecord.CardConfigJSON),
 		OutlineRequest: &req,
 	}
-	if err := s.enqueueAssist(job); err != nil {
+	if err := publishTemplateAssistTask(ctx, s.rabbitMQ, job); err != nil {
 		_ = repo.FailTask(context.Background(), taskID, templateAssistTaskTypeOutline, actor.TenantID, err.Error(), time.Now())
 		return nil, response.ErrServiceUnavailable(50314, "outline_queue_unavailable", "outline queue is busy, please retry")
 	}
@@ -363,21 +365,6 @@ func (s *TemplateService) GetOutlineTask(ctx context.Context, templateID int64,
 	return resp, nil
 }
 
-func (s *TemplateService) enqueueAssist(job assistJob) error {
-	select {
-	case s.assistJobs <- job:
-		return nil
-	default:
-		return fmt.Errorf("assist queue is full")
-	}
-}
-
-func (s *TemplateService) runAssistWorker() {
-	for job := range s.assistJobs {
-		s.executeAssist(context.Background(), job)
-	}
-}
-
 func (s *TemplateService) executeAssist(ctx context.Context, job assistJob) {
 	repo := repository.NewTemplateAssistTaskRepository(s.pool)
 	now := time.Now()
@@ -395,6 +382,28 @@ func (s *TemplateService) executeAssist(ctx context.Context, job assistJob) {
 	}
 }
 
+func (s *TemplateService) ProcessQueuedAssist(ctx context.Context, job assistJob) {
+	s.executeAssist(ctx, job)
+}
+
+func (s *TemplateService) ProcessQueuedAssistJob(ctx context.Context, job AssistJob) {
+	s.ProcessQueuedAssist(ctx, job)
+}
+
+func (s *TemplateService) IsAssistTaskTerminal(ctx context.Context, job AssistJob) (bool, error) {
+	record, err := repository.NewTemplateAssistTaskRepository(s.pool).GetTask(
+		ctx,
+		job.TaskID,
+		job.TaskType,
+		job.TenantID,
+		job.TemplateID,
+	)
+	if err != nil {
+		return false, err
+	}
+	return record.Status == "completed" || record.Status == "failed", nil
+}
+
 func (s *TemplateService) executeAnalyzeTask(ctx context.Context, repo repository.TemplateAssistTaskRepository, job assistJob) {
 	if job.AnalyzeRequest == nil {
 		s.failAssist(ctx, job.TaskID, job.TaskType, job.TenantID, fmt.Errorf("missing analyze request"))

server/internal/tenant/app/template_assist_queue.go

@@ -0,0 +1,44 @@
+package app
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
+)
+
+func publishTemplateAssistTask(ctx context.Context, client *rabbitmq.Client, job assistJob) error {
+	if client == nil {
+		return fmt.Errorf("rabbitmq client is not configured")
+	}
+
+	payload, err := json.Marshal(job)
+	if err != nil {
+		return fmt.Errorf("marshal template assist job: %w", err)
+	}
+
+	if err := client.PublishTemplateAssistTask(ctx, payload); err != nil {
+		return fmt.Errorf("publish template assist task: %w", err)
+	}
+	return nil
+}
+
+func decodeTemplateAssistTask(payload []byte) (*assistJob, error) {
+	var job assistJob
+	if err := json.Unmarshal(payload, &job); err != nil {
+		return nil, fmt.Errorf("decode template assist job: %w", err)
+	}
+	if job.TaskID == "" || job.TaskType == "" || job.TenantID <= 0 || job.TemplateID <= 0 {
+		return nil, fmt.Errorf("template assist job is incomplete")
+	}
+	return &job, nil
+}
+
+func DecodeTemplateAssistTask(payload []byte) (*AssistJob, error) {
+	job, err := decodeTemplateAssistTask(payload)
+	if err != nil {
+		return nil, err
+	}
+	return job, nil
+}

server/internal/tenant/app/template_service.go

@@ -13,6 +13,7 @@ import (
 	"github.com/geo-platform/tenant-api/internal/shared/auth"
 	"github.com/geo-platform/tenant-api/internal/shared/config"
 	"github.com/geo-platform/tenant-api/internal/shared/llm"
+	"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
 	"github.com/geo-platform/tenant-api/internal/shared/response"
 	"github.com/geo-platform/tenant-api/internal/shared/stream"
 	"github.com/geo-platform/tenant-api/internal/tenant/repository"
@@ -22,13 +23,12 @@ type TemplateService struct {
 	pool            *pgxpool.Pool
 	templates       repository.TemplateRepository
 	llm             llm.Client
+	rabbitMQ        *rabbitmq.Client
 	knowledge       *KnowledgeService
 	streams         *stream.GenerationHub
 	streamEnabled   bool
 	articleTimeout  time.Duration
 	maxOutputTokens int64
-	jobs            chan generationJob
-	assistJobs      chan assistJob
 }
 
 type generationJob struct {
@@ -49,21 +49,12 @@ func NewTemplateService(
 	pool *pgxpool.Pool,
 	templates repository.TemplateRepository,
 	llmClient llm.Client,
+	rabbitMQClient *rabbitmq.Client,
 	knowledge *KnowledgeService,
 	streams *stream.GenerationHub,
 	cfg config.GenerationConfig,
 	llmMaxOutputTokens int64,
 ) *TemplateService {
-	queueSize := cfg.QueueSize
-	if queueSize <= 0 {
-		queueSize = 128
-	}
-
-	workerCount := cfg.WorkerConcurrency
-	if workerCount <= 0 {
-		workerCount = 1
-	}
-
 	articleTimeout := cfg.ArticleTimeout
 	if articleTimeout <= 0 {
 		articleTimeout = 8 * time.Minute
@@ -73,18 +64,12 @@ func NewTemplateService(
 		pool:            pool,
 		templates:       templates,
 		llm:             llmClient,
+		rabbitMQ:        rabbitMQClient,
 		knowledge:       knowledge,
 		streams:         streams,
 		streamEnabled:   cfg.StreamEnabled,
 		articleTimeout:  articleTimeout,
 		maxOutputTokens: llmMaxOutputTokens,
-		jobs:            make(chan generationJob, queueSize),
-		assistJobs:      make(chan assistJob, queueSize),
-	}
-
-	for i := 0; i < workerCount; i++ {
-		go svc.runGenerationWorker()
-		go svc.runAssistWorker()
 	}
 
 	return svc
@@ -272,7 +257,13 @@ func (s *TemplateService) Generate(ctx context.Context, templateID int64, req Ge
 		return nil, response.ErrForbidden(40301, "quota_insufficient", "generation quota insufficient, please upgrade your plan")
 	}
 
-	inputJSON, _ := json.Marshal(req.InputParams)
+	taskInputParams := cloneInputParams(req.InputParams)
+	taskInputParams["enable_web_search"] = req.EnableWebSearch
+	if req.WebSearchLimit != nil {
+		taskInputParams["web_search_limit"] = *req.WebSearchLimit
+	}
+	taskInputParams = attachTemplateSnapshot(taskInputParams, templateRecord)
+	inputJSON, _ := json.Marshal(taskInputParams)
 
 	tx, err := s.pool.Begin(ctx)
 	if err != nil {
@@ -369,26 +360,22 @@ func (s *TemplateService) Generate(ctx context.Context, templateID int64, req Ge
 	}
 
 	job := generationJob{
-		OperatorID:     actor.UserID,
-		TenantID:       actor.TenantID,
-		ArticleID:      articleID,
-		TaskID:         taskID,
-		ReservationID:  reservationID,
-		TemplateKey:    templateRecord.TemplateKey,
-		TemplateName:   templateRecord.TemplateName,
-		PromptTemplate: templateRecord.PromptTemplate,
-		Params:         cloneInputParams(req.InputParams),
-		InitialTitle:   initialTitle,
-		WebSearch: llm.WebSearchOptions{
-			Enabled: req.EnableWebSearch,
-		},
-	}
-	if req.WebSearchLimit != nil {
-		job.WebSearch.Limit = *req.WebSearchLimit
+		OperatorID:    actor.UserID,
+		TenantID:      actor.TenantID,
+		ArticleID:     articleID,
+		TaskID:        taskID,
+		ReservationID: reservationID,
+		Params:        taskInputParams,
+		InitialTitle:  initialTitle,
 	}
 
-	if err := s.enqueueGeneration(job); err != nil {
-		s.failGeneration(context.Background(), job, job.InitialTitle, "queue_enqueue", err)
+	if err := publishArticleGenerationTask(ctx, s.rabbitMQ, articleGenerationTaskEnvelope{
+		TaskID:    taskID,
+		TaskType:  "template",
+		TenantID:  actor.TenantID,
+		ArticleID: articleID,
+	}); err != nil {
+		s.failGeneration(context.Background(), job, initialTitle, "queue_enqueue", err)
 		return nil, response.ErrServiceUnavailable(50310, "generation_queue_unavailable", "generation queue is busy, please retry")
 	}
 
@@ -427,21 +414,6 @@ func updateGeneratedArticleState(
 	return nil
 }
 
-func (s *TemplateService) enqueueGeneration(job generationJob) error {
-	select {
-	case s.jobs <- job:
-		return nil
-	default:
-		return fmt.Errorf("generation queue is full")
-	}
-}
-
-func (s *TemplateService) runGenerationWorker() {
-	for job := range s.jobs {
-		s.executeGeneration(context.Background(), job)
-	}
-}
-
 func (s *TemplateService) executeGeneration(ctx context.Context, job generationJob) {
 	now := time.Now()
 	title := strings.TrimSpace(job.InitialTitle)
@@ -622,6 +594,8 @@ func formatGenerationFailure(stage string, genErr error) string {
 
 func generationFailureStageLabel(stage string) string {
 	switch stage {
+	case "task_load":
+		return "加载生成任务失败"
 	case "queue_enqueue":
 		return "任务入队失败"
 	case "llm_generate":

server/internal/tenant/repository/audit_repo.go

@@ -70,9 +70,12 @@ func (r *auditRepository) CreateAuditLog(ctx context.Context, input AuditLogInpu
 }
 
 func (r *auditRepository) CreateGenerationTask(ctx context.Context, input GenerationTaskInput) (int64, error) {
-	operatorID := input.OperatorID
+	var operatorID *int64
+	if input.OperatorID > 0 {
+		operatorID = &input.OperatorID
+	}
 	return r.q.CreateGenerationTask(ctx, generated.CreateGenerationTaskParams{
-		OperatorID:         pgInt8(&operatorID),
+		OperatorID:         pgInt8(operatorID),
 		TenantID:           input.TenantID,
 		ArticleID:          pgInt8(input.ArticleID),
 		QuotaReservationID: pgInt8(input.QuotaReservationID),

server/internal/tenant/repository/quota_repo.go

@@ -57,10 +57,13 @@ func (r *quotaRepository) GetCurrentBalance(ctx context.Context, tenantID int64,
 }
 
 func (r *quotaRepository) InsertQuotaLedger(ctx context.Context, input QuotaLedgerInput) (int64, error) {
-	operatorID := input.OperatorID
+	var operatorID *int64
+	if input.OperatorID > 0 {
+		operatorID = &input.OperatorID
+	}
 	return r.q.InsertQuotaLedger(ctx, generated.InsertQuotaLedgerParams{
 		TenantID:      input.TenantID,
-		OperatorID:    pgInt8(&operatorID),
+		OperatorID:    pgInt8(operatorID),
 		QuotaType:     input.QuotaType,
 		Delta:         int32(input.Delta),
 		BalanceAfter:  int32(input.BalanceAfter),
@@ -72,10 +75,13 @@ func (r *quotaRepository) InsertQuotaLedger(ctx context.Context, input QuotaLedg
 
 func (r *quotaRepository) CreateQuotaReservation(ctx context.Context, input QuotaReservationInput) (int64, error) {
 	expireAt := input.ExpireAt
-	operatorID := input.OperatorID
+	var operatorID *int64
+	if input.OperatorID > 0 {
+		operatorID = &input.OperatorID
+	}
 	return r.q.CreateQuotaReservation(ctx, generated.CreateQuotaReservationParams{
 		TenantID:       input.TenantID,
-		OperatorID:     pgInt8(&operatorID),
+		OperatorID:     pgInt8(operatorID),
 		QuotaType:      input.QuotaType,
 		ResourceType:   input.ResourceType,
 		ResourceID:     pgInt8(input.ResourceID),

server/internal/tenant/transport/article_handler.go

@@ -51,6 +51,7 @@ func NewArticleHandler(a *bootstrap.App) *ArticleHandler {
 		promptGenerateSvc: app.NewPromptRuleGenerationService(
 			a.DB,
 			a.LLM,
+			a.RabbitMQ,
 			knowledgeSvc,
 			a.GenerationStreams,
 			a.Config.Generation,
@@ -369,6 +370,11 @@ func (h *ArticleHandler) Stream(c *gin.Context) {
 		return
 	}
 
+	fallback := buildGenerationSnapshot(detail)
+	if snapshot, ok := h.streams.Snapshot(id); !ok || (fallback.Done && snapshot.Status != fallback.Status) {
+		h.streams.Seed(fallback)
+	}
+
 	events, snapshot, cancel := h.streams.Subscribe(id)
 	defer cancel()
 
@@ -377,20 +383,10 @@ func (h *ArticleHandler) Stream(c *gin.Context) {
 			return
 		}
 		flusher.Flush()
+		if snapshot.Done {
+			return
+		}
 	} else {
-		fallback := stream.GenerationEvent{
-			Type:      "snapshot",
-			ArticleID: detail.ID,
-			Status:    detail.GenerateStatus,
-			Done:      detail.GenerateStatus == "completed" || detail.GenerateStatus == "failed",
-			UpdatedAt: time.Now().UTC(),
-		}
-		if detail.Title != nil {
-			fallback.Title = *detail.Title
-		}
-		if detail.MarkdownContent != nil {
-			fallback.Content = *detail.MarkdownContent
-		}
 		if err := writeSSE(c, fallback.Type, fallback); err != nil {
 			return
 		}
@@ -455,6 +451,32 @@ func writeSSE(c *gin.Context, event string, payload interface{}) error {
 	return nil
 }
 
+func buildGenerationSnapshot(detail *app.ArticleDetailResponse) stream.GenerationEvent {
+	event := stream.GenerationEvent{
+		Type:      "snapshot",
+		ArticleID: detail.ID,
+		Status:    detail.GenerateStatus,
+		Done:      detail.GenerateStatus == "completed" || detail.GenerateStatus == "failed",
+		UpdatedAt: time.Now().UTC(),
+	}
+	if detail.Title != nil {
+		event.Title = *detail.Title
+	}
+	if detail.MarkdownContent != nil {
+		event.Content = *detail.MarkdownContent
+	}
+	if detail.GenerateStatus == "completed" {
+		event.Type = "completed"
+	}
+	if detail.GenerateStatus == "failed" {
+		event.Type = "error"
+		if detail.GenerationErrorMessage != nil {
+			event.Error = *detail.GenerationErrorMessage
+		}
+	}
+	return event
+}
+
 func resolveOptimizeStreamErrorMessage(appErr *response.AppError) string {
 	if appErr == nil {
 		return "optimize selection failed"

server/internal/tenant/transport/template_handler.go

@@ -37,6 +37,7 @@ func NewTemplateHandler(a *bootstrap.App) *TemplateHandler {
 				a.Cache,
 			),
 			a.LLM,
+			a.RabbitMQ,
 			knowledgeSvc,
 			a.GenerationStreams,
 			a.Config.Generation,