feat: add generation job cancel and retry queue

admin-frontend/src/components/CreateStoryModal.vue

@@ -43,7 +43,10 @@ const outputMode = ref<'full_story' | 'storybook'>('full_story')
 const inputData = ref('')
 const educationTheme = ref('')
 const loading = ref(false)
+const canceling = ref(false)
+const cancelRequested = ref(false)
 const error = ref('')
+const activeGenerationJobId = ref<string | null>(null)
 
 // Data
 interface ChildProfile {
@@ -110,10 +113,17 @@ interface GenerationAcceptedResponse {
 
 interface GenerationJobDetail {
   story_id: number | null
+  status: string
+  current_step: string
   is_terminal: boolean
   error_message: string | null
 }
 
+interface GenerationJobActionResponse {
+  status: string
+  current_step: string
+}
+
 const JOB_POLL_INTERVAL_MS = 1500
 const JOB_POLL_MAX_ATTEMPTS = 80
 
@@ -121,6 +131,9 @@ const JOB_POLL_MAX_ATTEMPTS = 80
 function close() {
   emit('update:modelValue', false)
   error.value = ''
+  activeGenerationJobId.value = null
+  cancelRequested.value = false
+  canceling.value = false
 }
 
 function sleep(ms: number) {
@@ -132,6 +145,9 @@ function sleep(ms: number) {
 async function waitForStoryId(jobId: string) {
   for (let attempt = 0; attempt < JOB_POLL_MAX_ATTEMPTS; attempt += 1) {
     const detail = await api.get<GenerationJobDetail>(`/api/generations/jobs/${jobId}`)
+    if (detail.status === 'canceled' || detail.current_step === 'generation_canceled') {
+      return null
+    }
     if (detail.story_id) {
       return detail.story_id
     }
@@ -144,6 +160,27 @@ async function waitForStoryId(jobId: string) {
   throw new Error('任务已提交，但主内容落库超时，请稍后到故事库查看最新结果')
 }
 
+async function cancelGenerationJob() {
+  if (!activeGenerationJobId.value || canceling.value || cancelRequested.value) return
+
+  canceling.value = true
+  error.value = ''
+  try {
+    const result = await api.post<GenerationJobActionResponse>(
+      `/api/generations/jobs/${activeGenerationJobId.value}/cancel`,
+    )
+    cancelRequested.value = true
+    if (result.status === 'canceled' || result.current_step === 'generation_canceled') {
+      loading.value = false
+      close()
+    }
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '取消任务失败'
+  } finally {
+    canceling.value = false
+  }
+}
+
 async function fetchProfiles() {
   if (!userStore.user) return
   profileError.value = ''
@@ -192,6 +229,8 @@ async function generateStory() {
   }
   
   loading.value = true
+  cancelRequested.value = false
+  activeGenerationJobId.value = null
   error.value = ''
 
   try {
@@ -211,8 +250,13 @@ async function generateStory() {
     if (!jobId) {
       throw new Error('生成任务已创建，但缺少任务编号')
     }
+    activeGenerationJobId.value = jobId
 
     const storyId = accepted.id ?? await waitForStoryId(jobId)
+    if (storyId === null) {
+      close()
+      return
+    }
     close()
     if (requestedOutputMode.value === 'storybook') {
       router.push(`/storybook/view/${storyId}`)
@@ -223,6 +267,8 @@ async function generateStory() {
     error.value = e instanceof Error ? e.message : '生成失败'
   } finally {
     loading.value = false
+    activeGenerationJobId.value = null
+    cancelRequested.value = false
   }
 }
 </script>
@@ -253,6 +299,22 @@ async function generateStory() {
           :title="generationTitle"
           :steps="generationSteps"
         />
+        <div
+          v-if="loading && activeGenerationJobId"
+          class="fixed bottom-10 z-[110] flex flex-col items-center gap-3"
+        >
+          <BaseButton
+            variant="secondary"
+            :loading="canceling"
+            :disabled="cancelRequested"
+            @click="cancelGenerationJob"
+          >
+            {{ cancelRequested ? '正在取消任务...' : '取消任务' }}
+          </BaseButton>
+          <p class="text-sm text-white/70">
+            {{ cancelRequested ? '已提交取消请求，会在安全检查点停止任务。' : '如果是误触发起，可以现在取消后台任务。' }}
+          </p>
+        </div>
 
         <!-- 模态框内容 -->
         <div v-else class="relative w-full max-w-2xl max-h-[90vh] overflow-y-auto bg-[#1C2035] border border-gray-700/50 rounded-3xl shadow-2xl p-6 md:p-8">

admin-frontend/src/components/GenerationTrace.vue

@@ -14,6 +14,8 @@ interface GenerationJobSummary {
   progress_percent: number
   progress_label: string
   is_terminal: boolean
+  can_cancel: boolean
+  can_retry: boolean
   result_snapshot: Record<string, unknown>
   error_message: string | null
   created_at: string
@@ -63,6 +65,7 @@ const jobs = ref<GenerationJobSummary[]>([])
 const activeJob = ref<GenerationJobDetail | null>(null)
 const providerStats = ref<GenerationProviderStats | null>(null)
 const loading = ref(false)
+const actionLoading = ref(false)
 const error = ref('')
 let refreshTimer: ReturnType<typeof setInterval> | null = null
 
@@ -94,6 +97,7 @@ const statusClassMap: Record<string, string> = {
   succeeded: 'border-emerald-200 bg-emerald-50 text-emerald-700',
   completed: 'border-emerald-200 bg-emerald-50 text-emerald-700',
   degraded_completed: 'border-orange-200 bg-orange-50 text-orange-700',
+  canceled: 'border-slate-200 bg-slate-100 text-slate-700',
   failed: 'border-rose-200 bg-rose-50 text-rose-700',
 }
 
@@ -108,6 +112,7 @@ function statusLabel(status?: string) {
     succeeded: '成功',
     completed: '已完成',
     degraded_completed: '降级完成',
+    canceled: '已取消',
     failed: '失败',
   }
   return labels[status ?? ''] ?? '未知'
@@ -117,6 +122,8 @@ function eventLabel(eventType: string) {
   const labels: Record<string, string> = {
     request_accepted: '请求接收',
     worker_started: '后台任务开始',
+    retry_queued: '重新排队',
+    cancel_requested: '已请求取消',
     context_prepared: '上下文准备',
     narrative_generated: '正文生成',
     story_saved: '故事保存',
@@ -137,6 +144,7 @@ function eventLabel(eventType: string) {
     asset_retry_started: '资源重试开始',
     asset_retry_completed: '资源重试完成',
     asset_retry_failed: '资源重试失败',
+    generation_canceled: '任务已取消',
     generation_completed: '生成完成',
     generation_failed: '生成失败',
   }
@@ -213,6 +221,36 @@ async function refresh() {
   }
 }
 
+async function cancelActiveJob() {
+  if (!activeJob.value || actionLoading.value) return
+
+  actionLoading.value = true
+  error.value = ''
+  try {
+    await api.post(`/api/generations/jobs/${activeJob.value.id}/cancel`)
+    await refresh()
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '取消任务失败'
+  } finally {
+    actionLoading.value = false
+  }
+}
+
+async function retryActiveJob() {
+  if (!activeJob.value || actionLoading.value) return
+
+  actionLoading.value = true
+  error.value = ''
+  try {
+    await api.post(`/api/generations/jobs/${activeJob.value.id}/retry`)
+    await refresh()
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '重新排队失败'
+  } finally {
+    actionLoading.value = false
+  }
+}
+
 function stopAutoRefresh() {
   if (refreshTimer) {
     clearInterval(refreshTimer)
@@ -334,10 +372,30 @@ defineExpose({ refresh })
                 当前步骤：{{ eventLabel(activeJob.current_step) }}
               </div>
             </div>
+            <div class="flex flex-wrap items-center justify-end gap-2">
+              <button
+                v-if="activeJob.can_cancel"
+                type="button"
+                class="rounded-full border border-amber-200 bg-amber-50 px-3 py-1 text-xs font-medium text-amber-700 transition hover:bg-amber-100 disabled:cursor-not-allowed disabled:opacity-60"
+                :disabled="actionLoading"
+                @click="cancelActiveJob"
+              >
+                {{ actionLoading ? '处理中...' : '取消任务' }}
+              </button>
+              <button
+                v-if="activeJob.can_retry"
+                type="button"
+                class="rounded-full border border-sky-200 bg-sky-50 px-3 py-1 text-xs font-medium text-sky-700 transition hover:bg-sky-100 disabled:cursor-not-allowed disabled:opacity-60"
+                :disabled="actionLoading"
+                @click="retryActiveJob"
+              >
+                {{ actionLoading ? '处理中...' : '重新排队' }}
+              </button>
               <span class="rounded-full border px-3 py-1 text-xs font-medium" :class="statusClass(activeJob.status)">
                 {{ statusLabel(activeJob.status) }}
               </span>
             </div>
+          </div>
 
           <div>
             <div class="mb-1 flex items-center justify-between text-xs" :class="mutedClass">

backend/app/api/stories.py

@@ -40,6 +40,7 @@ from app.services.generation_jobs import (
     get_user_generation_ops_summary,
     get_user_provider_analytics,
     list_story_generation_jobs,
+    request_generation_job_cancel,
 )
 from app.services.memory_service import build_enhanced_memory_context
 from app.services.provider_router import (
@@ -88,6 +89,32 @@ async def get_generation_job(
     return await get_generation_job_detail(db, job_id=job_id, user_id=user.id)
 
 
+@router.post(
+    "/generations/jobs/{job_id}/cancel",
+    response_model=GenerationJobSummaryResponse,
+)
+async def cancel_generation_job(
+    job_id: str,
+    user: User = Depends(require_user),
+    db: AsyncSession = Depends(get_db),
+):
+    """Request cancellation for one queued/running generation job."""
+    return await request_generation_job_cancel(db, job_id=job_id, user_id=user.id)
+
+
+@router.post(
+    "/generations/jobs/{job_id}/retry",
+    response_model=GenerationJobSummaryResponse,
+)
+async def retry_generation_job(
+    job_id: str,
+    user: User = Depends(require_user),
+    db: AsyncSession = Depends(get_db),
+):
+    """Queue one new generation job from a failed/canceled terminal job."""
+    return await story_service.retry_generation_job_service(job_id, user.id, db)
+
+
 @router.get(
     "/generations/ops-summary",
     response_model=GenerationOpsSummaryResponse,

backend/app/schemas/story_schemas.py

@@ -195,6 +195,8 @@ class GenerationJobSummaryResponse(BaseModel):
     progress_percent: int
     progress_label: str
     is_terminal: bool
+    can_cancel: bool = False
+    can_retry: bool = False
     result_snapshot: dict[str, Any] = Field(default_factory=dict)
     error_message: str | None = None
     created_at: datetime

backend/app/services/generation_jobs.py

@@ -16,6 +16,26 @@ from app.db.models import GenerationJob, GenerationJobEvent, Story
 logger = get_logger(__name__)
 
 
+def _is_terminal_status(status: str) -> bool:
+    return status in {"completed", "degraded_completed", "failed", "canceled"}
+
+
+def _job_supports_queue_control(job: GenerationJob) -> bool:
+    return job.output_mode in {"story", "storybook"}
+
+
+def generation_job_can_cancel(job: GenerationJob) -> bool:
+    return (
+        _job_supports_queue_control(job)
+        and job.status == "running"
+        and job.current_step != "cancel_requested"
+    )
+
+
+def generation_job_can_retry(job: GenerationJob) -> bool:
+    return _job_supports_queue_control(job) and job.status in {"failed", "canceled"}
+
+
 def _story_snapshot(story: Story | None) -> dict[str, Any]:
     if story is None:
         return {}
@@ -50,6 +70,13 @@ def _job_progress(job: GenerationJob) -> dict[str, Any]:
             "is_terminal": True,
         }
 
+    if job.status == "canceled":
+        return {
+            "progress_percent": 100,
+            "progress_label": "已取消",
+            "is_terminal": True,
+        }
+
     if job.status in {"completed", "degraded_completed"}:
         return {
             "progress_percent": 100,
@@ -59,7 +86,9 @@ def _job_progress(job: GenerationJob) -> dict[str, Any]:
 
     progress_map: dict[str, tuple[int, str]] = {
         "request_accepted": (5, "已接收请求"),
+        "retry_queued": (8, "重新排队中"),
         "worker_started": (12, "后台任务已开始"),
+        "cancel_requested": (15, "已请求取消"),
         "context_prepared": (20, "上下文已准备"),
         "narrative_generated": (45, "正文已生成"),
         "story_saved": (60, "主记录已保存"),
@@ -83,6 +112,7 @@ def _job_progress(job: GenerationJob) -> dict[str, Any]:
         "postprocessing_queued": (90, "后处理已排队"),
         "asset_generation_completed": (100, "资源已完成"),
         "asset_retry_completed": (100, "资源重试完成"),
+        "generation_canceled": (100, "任务已取消"),
         "generation_completed": (100, "生成完成"),
         "generation_stale_failed": (100, "任务超时已收敛"),
     }
@@ -106,6 +136,8 @@ def _is_stale_job(job: GenerationJob, *, stale_after_minutes: int) -> bool:
 
 
 def _failure_label(job: GenerationJob) -> str:
+    if job.status == "canceled":
+        return "任务已取消"
     if job.current_step == "generation_stale_failed":
         return "任务超时"
     if job.output_mode == "asset_retry":
@@ -196,7 +228,7 @@ async def claim_generation_job_for_worker(
         .where(
             GenerationJob.id == job_id,
             GenerationJob.status == "running",
-            GenerationJob.current_step == "request_accepted",
+            GenerationJob.current_step.in_(["request_accepted", "retry_queued"]),
         )
         .values(current_step="worker_started")
     )
@@ -283,6 +315,8 @@ def generation_job_to_summary(job: GenerationJob) -> dict[str, Any]:
         "status": job.status,
         "current_step": job.current_step,
         **progress,
+        "can_cancel": generation_job_can_cancel(job),
+        "can_retry": generation_job_can_retry(job),
         "result_snapshot": job.result_snapshot or {},
         "error_message": job.error_message,
         "created_at": job.created_at,
@@ -290,6 +324,88 @@ def generation_job_to_summary(job: GenerationJob) -> dict[str, Any]:
     }
 
 
+async def get_generation_job_for_user(
+    db: AsyncSession,
+    *,
+    job_id: str,
+    user_id: str,
+) -> GenerationJob:
+    """Load one generation job owned by the current user."""
+
+    result = await db.execute(
+        select(GenerationJob).where(
+            GenerationJob.id == job_id,
+            GenerationJob.user_id == user_id,
+        )
+    )
+    job = result.scalar_one_or_none()
+    if job is None:
+        raise HTTPException(status_code=404, detail="Generation job not found")
+    return job
+
+
+async def request_generation_job_cancel(
+    db: AsyncSession,
+    *,
+    job_id: str,
+    user_id: str,
+) -> dict[str, Any]:
+    """Request cancellation for one queued/running generation job."""
+
+    job = await get_generation_job_for_user(db, job_id=job_id, user_id=user_id)
+
+    if not _job_supports_queue_control(job):
+        raise HTTPException(status_code=409, detail="当前任务不支持取消")
+
+    if job.status == "canceled":
+        return generation_job_to_summary(job)
+
+    if _is_terminal_status(job.status):
+        raise HTTPException(status_code=409, detail="当前任务已终止，无法取消")
+
+    if job.current_step == "cancel_requested":
+        return generation_job_to_summary(job)
+
+    if job.current_step in {"request_accepted", "retry_queued"}:
+        story = None
+        if job.story_id is not None:
+            story = (
+                await db.execute(
+                    select(Story).where(
+                        Story.id == job.story_id,
+                        Story.user_id == job.user_id,
+                    )
+                )
+            ).scalar_one_or_none()
+
+        await finish_generation_job(
+            db,
+            job=job,
+            story=story,
+            status="canceled",
+            current_step="generation_canceled",
+            error_message="Generation canceled by user before worker execution started.",
+            message="Generation job was canceled before worker execution started.",
+        )
+        return generation_job_to_summary(job)
+
+    previous_step = job.current_step
+    job.error_message = "Cancellation requested by user."
+    await record_generation_event(
+        db,
+        job=job,
+        story_id=job.story_id,
+        event_type="cancel_requested",
+        status="running",
+        message="Cancellation requested; worker will stop at the next safe checkpoint.",
+        metadata={"requested_from_step": previous_step},
+        commit=False,
+    )
+    await db.commit()
+    await db.refresh(job)
+    return generation_job_to_summary(job)
+
+
 async def get_generation_job_detail(
     db: AsyncSession,
     *,

backend/app/services/story_service.py

@@ -37,6 +37,9 @@ from app.services.generation_jobs import (
     create_generation_job,
     ensure_no_active_story_generation_job,
     finish_generation_job,
+    generation_job_can_retry,
+    generation_job_to_summary,
+    get_generation_job_for_user,
     record_generation_event,
 )
 from app.services.memory_service import build_enhanced_memory_context
@@ -73,6 +76,10 @@ class AssetCompletionResult:
         return self.status == StoryAssetStatus.READY and self.error is None
 
 
+class GenerationJobCanceledError(Exception):
+    """Raised when a running worker job has been canceled by the user."""
+
+
 async def _record_job_event_if_present(
     db: AsyncSession,
     *,
@@ -99,6 +106,33 @@ async def _record_job_event_if_present(
     )
 
 
+async def _stop_if_job_cancel_requested(
+    db: AsyncSession,
+    *,
+    job,
+    story: Story | None = None,
+) -> None:
+    """Stop a worker-owned job at the next safe checkpoint after cancellation."""
+
+    if job is None:
+        return
+
+    await db.refresh(job)
+    if job.current_step != "cancel_requested":
+        return
+
+    await finish_generation_job(
+        db,
+        job=job,
+        story=story,
+        status="canceled",
+        current_step="generation_canceled",
+        error_message="Generation canceled by user.",
+        message="Generation job was canceled after a user request.",
+    )
+    raise GenerationJobCanceledError()
+
+
 def _asset_result_metadata(result: AssetCompletionResult) -> dict:
     """Build JSON-safe metadata for asset workflow events."""
 
@@ -192,6 +226,7 @@ async def _prepare_generation_context(
             "has_memory_context": bool(memory_context),
         },
     )
+    await _stop_if_job_cancel_requested(db, job=job)
     return resolved_profile_id, resolved_universe_id, memory_context
 
 
@@ -318,6 +353,7 @@ async def _generate_storybook_image_assets(
     ]
 
     logger.info("storybook_parallel_generation_start", page_count=len(storybook.pages))
+    await _stop_if_job_cancel_requested(db, job=job)
     await _record_job_event_if_present(
         db,
         job=job,
@@ -334,6 +370,7 @@ async def _generate_storybook_image_assets(
         nonlocal cover_failed
 
         if storybook.cover_prompt and not storybook.cover_url:
+            await _stop_if_job_cancel_requested(db, job=job)
             try:
                 return await generate_image(
                     storybook.cover_prompt,
@@ -350,6 +387,7 @@ async def _generate_storybook_image_assets(
         if not page.image_prompt or page.image_url:
             return
 
+        await _stop_if_job_cancel_requested(db, job=job)
         try:
             page.image_url = await generate_image(
                 page.image_prompt,
@@ -506,6 +544,7 @@ async def _complete_cover_image_asset(
 
     sync_story_status(story, image_status=StoryAssetStatus.GENERATING)
     await db.commit()
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
     await _record_job_event_if_present(
         db,
         job=job,
@@ -517,6 +556,7 @@ async def _complete_cover_image_asset(
     )
 
     try:
+        await _stop_if_job_cancel_requested(db, job=job, story=story)
         image_url = await generate_image(
             story.cover_prompt,
             db=db,
@@ -605,6 +645,7 @@ async def _complete_storybook_image_assets(
 
     sync_story_status(story, image_status=StoryAssetStatus.GENERATING)
     await db.commit()
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
     await _record_job_event_if_present(
         db,
         job=job,
@@ -620,6 +661,7 @@ async def _complete_storybook_image_assets(
     completed_pages: list[int] = []
 
     if story.cover_prompt and not story.image_url:
+        await _stop_if_job_cancel_requested(db, job=job, story=story)
         try:
             story.image_url = await generate_image(
                 story.cover_prompt,
@@ -658,6 +700,7 @@ async def _complete_storybook_image_assets(
         if not page.get("image_prompt") or page.get("image_url"):
             continue
 
+        await _stop_if_job_cancel_requested(db, job=job, story=story)
         try:
             page["image_url"] = await generate_image(
                 page["image_prompt"],
@@ -800,6 +843,7 @@ async def _complete_audio_asset(
 
     sync_story_status(story, audio_status=StoryAssetStatus.GENERATING)
     await db.commit()
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
     await _record_job_event_if_present(
         db,
         job=job,
@@ -811,6 +855,7 @@ async def _complete_audio_asset(
     )
 
     try:
+        await _stop_if_job_cancel_requested(db, job=job, story=story)
         audio_data = await text_to_speech(
             story.story_text,
             db=db,
@@ -933,6 +978,7 @@ async def generate_and_save_story(
     )
 
     try:
+        await _stop_if_job_cancel_requested(db, job=job)
         result = await generate_story_content(
             input_type=request.type,
             data=request.data,
@@ -955,8 +1001,9 @@ async def generate_and_save_story(
         message="Story narrative was generated.",
         metadata={"mode": result.mode, "title": result.title},
     )
+    await _stop_if_job_cancel_requested(db, job=job)
 
-    return await _persist_text_story_result(
+    story = await _persist_text_story_result(
         result=result,
         user_id=user_id,
         profile_id=profile_id,
@@ -964,6 +1011,8 @@ async def generate_and_save_story(
         db=db,
         job=job,
     )
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
+    return story
 
 
 async def generate_full_story_service(
@@ -975,6 +1024,7 @@ async def generate_full_story_service(
 ) -> FullStoryResponse:
     """Generate story with parallel image generation."""
     story = await generate_and_save_story(request, user_id, db, job=job)
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
     image_url: str | None = None
     errors: dict[str, str | None] = {}
 
@@ -1036,6 +1086,7 @@ async def generate_storybook_service(
     )
 
     try:
+        await _stop_if_job_cancel_requested(db, job=job)
         storybook = await generate_storybook(
             keywords=request.keywords,
             page_count=request.page_count,
@@ -1060,12 +1111,14 @@ async def generate_storybook_service(
             "page_count": len(storybook.pages),
         },
     )
+    await _stop_if_job_cancel_requested(db, job=job)
 
     final_cover_url = storybook.cover_url
     cover_failed = False
     failed_pages: list[int] = []
 
     if request.generate_images:
+        await _stop_if_job_cancel_requested(db, job=job)
         (
             final_cover_url,
             cover_failed,
@@ -1089,6 +1142,7 @@ async def generate_storybook_service(
         db=db,
         job=job,
     )
+    await _stop_if_job_cancel_requested(db, job=job, story=story)
 
     response_pages = _storybook_pages_to_response(pages_data)
 
@@ -1124,6 +1178,18 @@ async def generate_generation_service(
         request_payload=request.model_dump(mode="json"),
     )
 
+    await _dispatch_generation_job(db, job=job)
+
+    return _build_queued_generation_response(request, job_id=job.id)
+
+
+async def _dispatch_generation_job(
+    db: AsyncSession,
+    *,
+    job: GenerationJob,
+) -> None:
+    """Dispatch one accepted generation job to the background worker."""
+
     try:
         from app.tasks.generation_workflow import run_generation_workflow_task
 
@@ -1144,8 +1210,6 @@ async def generate_generation_service(
             detail="后台生成任务派发失败，请确认 worker 可用后重试。",
         ) from exc
 
-    return _build_queued_generation_response(request, job_id=job.id)
-
 
 def _build_queued_generation_response(
     request: GenerationRequest,
@@ -1184,6 +1248,8 @@ async def execute_generation_job_service(
             db,
             job=job,
         )
+    except GenerationJobCanceledError:
+        return _build_canceled_generation_response(job)
     except HTTPException as exc:
         await finish_generation_job(
             db,
@@ -1210,6 +1276,24 @@ async def execute_generation_job_service(
     return response
 
 
+def _build_canceled_generation_response(job: GenerationJob) -> GenerationResponse:
+    """Build a compact response for a worker job that ended as canceled."""
+
+    snapshot = job.result_snapshot or {}
+    return GenerationResponse(
+        id=snapshot.get("story_id"),
+        generation_job_id=job.id,
+        title="生成任务已取消",
+        mode="storybook" if job.output_mode == "storybook" else "generated",
+        generation_status=str(snapshot.get("generation_status") or "failed"),
+        text_status=str(snapshot.get("text_status") or "failed"),
+        image_status=str(snapshot.get("image_status") or "not_requested"),
+        audio_status=str(snapshot.get("audio_status") or "not_requested"),
+        last_error=str(snapshot.get("last_error") or "Generation canceled by user."),
+        retryable_assets=list(snapshot.get("retryable_assets") or []),
+    )
+
+
 async def run_generation_job_service(
     job_id: str,
     db: AsyncSession,
@@ -1225,6 +1309,46 @@ async def run_generation_job_service(
     return job
 
 
+async def retry_generation_job_service(
+    job_id: str,
+    user_id: str,
+    db: AsyncSession,
+) -> dict:
+    """Clone one failed/canceled generation job and queue it again."""
+
+    source_job = await get_generation_job_for_user(db, job_id=job_id, user_id=user_id)
+    if not generation_job_can_retry(source_job):
+        raise HTTPException(status_code=409, detail="当前任务还不能重新排队")
+
+    if source_job.story_id is not None:
+        await ensure_no_active_story_generation_job(
+            db,
+            story_id=source_job.story_id,
+            user_id=user_id,
+        )
+
+    retry_job = await create_generation_job(
+        db,
+        user_id=user_id,
+        output_mode=source_job.output_mode,
+        input_type=source_job.input_type,
+        request_payload=source_job.request_payload or {},
+        story_id=source_job.story_id,
+    )
+    await record_generation_event(
+        db,
+        job=retry_job,
+        story_id=retry_job.story_id,
+        event_type="retry_queued",
+        status="queued",
+        message="Retry job accepted from a previous terminal generation.",
+        metadata={"source_job_id": source_job.id},
+    )
+    await _dispatch_generation_job(db, job=retry_job)
+    await db.refresh(retry_job)
+    return generation_job_to_summary(retry_job)
+
+
 async def _generate_generation_service_with_job(
     request: GenerationRequest,
     user_id: str,

backend/tests/test_generation_jobs.py

@@ -15,6 +15,8 @@ from app.services.adapters.storybook.primary import Storybook, StorybookPage
 from app.services.adapters.text.models import StoryOutput
 from app.services.generation_jobs import (
     create_generation_job,
+    finish_generation_job,
+    get_generation_job_detail,
     mark_stale_generation_jobs,
     record_generation_event,
 )
@@ -847,3 +849,187 @@ async def test_retry_assets_rejects_when_story_has_active_job(
         assert "已有运行中的任务" in response.json()["detail"]
     finally:
         app.dependency_overrides.clear()
+
+
+async def test_cancel_queued_generation_job_marks_it_canceled(
+    db_session,
+    auth_token,
+    test_user,
+):
+    async def override_get_db():
+        yield db_session
+
+    app.dependency_overrides[get_db] = override_get_db
+    transport = ASGITransport(app=app)
+    task_delay_path = "app.tasks.generation_workflow.run_generation_workflow_task.delay"
+
+    try:
+        with patch(task_delay_path) as mock_delay:
+            async with AsyncClient(transport=transport, base_url="http://test") as client:
+                client.cookies.set("access_token", auth_token)
+
+                response = await client.post(
+                    "/api/generations",
+                    json={
+                        "output_mode": "story",
+                        "type": "keywords",
+                        "data": "小狐狸, 月亮船",
+                        "generate_images": False,
+                    },
+                )
+
+                assert response.status_code == 202
+                job_id = response.json()["generation_job_id"]
+                mock_delay.assert_called_once_with(job_id)
+
+                cancel_response = await client.post(f"/api/generations/jobs/{job_id}/cancel")
+
+            assert cancel_response.status_code == 200
+            canceled_job = cancel_response.json()
+            assert canceled_job["status"] == "canceled"
+            assert canceled_job["current_step"] == "generation_canceled"
+            assert canceled_job["can_cancel"] is False
+            assert canceled_job["can_retry"] is True
+
+            detail = await get_generation_job_detail(
+                db_session,
+                job_id=job_id,
+                user_id=test_user.id,
+            )
+            assert [event["event_type"] for event in detail["events"]] == [
+                "request_accepted",
+                "generation_canceled",
+            ]
+    finally:
+        app.dependency_overrides.clear()
+
+
+async def test_cancel_running_generation_job_marks_cancel_requested(
+    db_session,
+    auth_token,
+    test_user,
+):
+    async def override_get_db():
+        yield db_session
+
+    app.dependency_overrides[get_db] = override_get_db
+    transport = ASGITransport(app=app)
+
+    job = await create_generation_job(
+        db_session,
+        user_id=test_user.id,
+        output_mode="story",
+        input_type="keywords",
+        request_payload={
+            "output_mode": "story",
+            "type": "keywords",
+            "data": "小熊, 森林",
+            "generate_images": False,
+        },
+    )
+    await record_generation_event(
+        db_session,
+        job=job,
+        event_type="worker_started",
+        status="running",
+        message="Generation worker started processing the accepted request.",
+    )
+
+    try:
+        async with AsyncClient(transport=transport, base_url="http://test") as client:
+            client.cookies.set("access_token", auth_token)
+
+            response = await client.post(f"/api/generations/jobs/{job.id}/cancel")
+
+        assert response.status_code == 200
+        data = response.json()
+        assert data["status"] == "running"
+        assert data["current_step"] == "cancel_requested"
+        assert data["can_cancel"] is False
+        assert data["can_retry"] is False
+
+        refreshed_job = (
+            await db_session.execute(select(GenerationJob).where(GenerationJob.id == job.id))
+        ).scalar_one()
+        assert refreshed_job.current_step == "cancel_requested"
+        assert refreshed_job.error_message == "Cancellation requested by user."
+    finally:
+        app.dependency_overrides.clear()
+
+
+async def test_retry_failed_generation_job_requeues_new_worker_job(
+    db_session,
+    auth_token,
+    test_user,
+    mock_text_provider,
+):
+    async def override_get_db():
+        yield db_session
+
+    app.dependency_overrides[get_db] = override_get_db
+    transport = ASGITransport(app=app)
+    task_delay_path = "app.tasks.generation_workflow.run_generation_workflow_task.delay"
+
+    failed_job = await create_generation_job(
+        db_session,
+        user_id=test_user.id,
+        output_mode="story",
+        input_type="keywords",
+        request_payload={
+            "output_mode": "story",
+            "type": "keywords",
+            "data": "小鹿, 星星",
+            "generate_images": False,
+        },
+    )
+    await finish_generation_job(
+        db_session,
+        job=failed_job,
+        story=None,
+        status="failed",
+        current_step="generation_failed",
+        error_message="upstream timeout",
+        message="Generation failed before a durable story result was available.",
+    )
+
+    try:
+        with patch(task_delay_path) as mock_delay:
+            async with AsyncClient(transport=transport, base_url="http://test") as client:
+                client.cookies.set("access_token", auth_token)
+
+                response = await client.post(f"/api/generations/jobs/{failed_job.id}/retry")
+
+            assert response.status_code == 200
+            data = response.json()
+            assert data["id"] != failed_job.id
+            assert data["status"] == "running"
+            assert data["current_step"] == "retry_queued"
+            assert data["can_cancel"] is True
+            assert data["can_retry"] is False
+            mock_delay.assert_called_once_with(data["id"])
+
+        retried_job_id = data["id"]
+        await run_generation_job_service(retried_job_id, db_session)
+
+        retried_job = (
+            await db_session.execute(
+                select(GenerationJob).where(GenerationJob.id == retried_job_id)
+            )
+        ).scalar_one()
+        assert retried_job.status == "completed"
+        assert retried_job.current_step == "generation_completed"
+
+        events = (
+            await db_session.execute(
+                select(GenerationJobEvent)
+                .where(GenerationJobEvent.job_id == retried_job_id)
+                .order_by(GenerationJobEvent.id)
+            )
+        ).scalars().all()
+        assert [event.event_type for event in events[:3]] == [
+            "request_accepted",
+            "retry_queued",
+            "worker_started",
+        ]
+    finally:
+        app.dependency_overrides.clear()

docs/planning/demo-checklist.md

@@ -126,7 +126,7 @@ DreamWeaver 是面向 3-8 岁亲子场景的个性化 AI 绘本与陪伴式讲
 
 ### 2:20 - 3:00 取舍与下一步
 
-求职版优先稳定闭环和可解释性，不做支付、多租户和复杂监控。现在 job/event 已能查询 workflow、资产补全、provider 调用轨迹和聚合指标，用户端和管理端也能展示生成轨迹与跨故事 Provider 运营摘要；下一步会迁移到后台 worker。
+求职版优先稳定闭环和可解释性，不做支付、多租户和复杂监控。现在 job/event 已能查询 workflow、资产补全、provider 调用轨迹和聚合指标，用户端和管理端也能展示生成轨迹与跨故事 Provider 运营摘要；统一生成也已经迁移到后台 worker，下一步是补取消/重试队列。
 
 ---
 

docs/planning/demo-package.md

@@ -51,7 +51,7 @@ SMOKE_AUDIO=1 ./scripts/demo_smoke.sh
 - **AI 不确定性处理**：主内容和资产拆开，图片/音频失败不阻塞阅读。
 - **Provider 产品化**：用户看到稳定能力，系统内部用 Capability / Provider / Adapter / Routing Policy 管供应链。
 - **可观测性**：generation job/event 让生成过程、失败恢复和 Provider 成本可解释。
-- **可继续生产化**：前端已有轮询形态，后端已有任务事件模型，下一步可以迁移到 worker。
+- **可继续生产化**：统一生成已经迁移到 worker，前端轮询和任务事件模型也已打通，下一步是补取消/重试队列和更完整监控。
 
 ---
 
@@ -63,4 +63,4 @@ SMOKE_AUDIO=1 ./scripts/demo_smoke.sh
 | 图片生成失败 | 展示 `degraded_completed` 与资源重试 |
 | Docker 冷启动慢 | 演示前先跑 smoke 并保持容器运行 |
 | Provider 追问过深 | 回到 Capability / Provider / Adapter / Routing Policy 四层解释 |
-| 生产化追问 | 说明下一步是 worker 化、监控告警、密钥治理和 Provider analytics 扩展 |
+| 生产化追问 | 说明下一步是取消/重试队列、监控告警、密钥治理和 Provider analytics 扩展 |

docs/planning/interview-pitch.md

@@ -83,4 +83,4 @@ AI 生成产品最大的问题不是“能不能调模型”，而是结果不
 
 ### 这个项目下一步怎么上线？
 
-我会先把当前轻量 job/event 模型迁移到后台 worker 和进度轮询，再补跨时间窗口的 provider 运营分析。生产上线前还需要补真实用户鉴权配置、密钥管理、监控告警和部署策略。
+我已经把当前轻量 job/event 模型迁移到后台 worker，并打通了前端进度轮询；下一步会补取消/重试队列，再继续扩展跨时间窗口和跨用户维度的 provider 运营分析。生产上线前还需要补真实用户鉴权配置、密钥管理、监控告警和部署策略。

docs/planning/week-2-to-4-execution-backlog.md

@@ -71,6 +71,7 @@ Week 2 已完成演示闭环、统一生成工作流、generation job/event、
 | W4-08 | Ops | 任务运行概览与失败摘要 | `GET /api/generations/ops-summary` + 最近失败列表 | P1 | Done |
 | W4-09 | Workflow | 卡住任务自动收敛 | `GENERATION_JOB_STALE_MINUTES` + Celery beat stale job maintenance | P1 | Done |
 | W4-10 | Workflow | 防止重复资产任务 | 运行中故事拒绝重复封面/音频/资产重试请求 | P1 | Done |
+| W4-11 | Workflow | 生成任务取消与重新排队 | 取消已提交任务，失败/取消任务可重新排队 | P1 | Done |
 
 ---
 

docs/planning/week-4-sprint-review.md

@@ -18,6 +18,11 @@
   - 时间线能展示阅读记录与记忆沉淀
 - Week 4 已补齐绘本阅读位置恢复。
 - Week 4 已输出架构说明和 Demo 包装文档。
+- 生产化主线已继续推进：
+  - `POST /api/generations` 已迁移到后台 worker
+  - 创建弹窗会先拿到 `generation_job_id`，再轮询主记录落库
+  - 统一生成链路的 smoke、测试和前端构建已跟进到异步语义
+  - 首版取消/重试队列已落地，支持取消已提交任务和从失败/取消任务重新排队
 
 ---
 
@@ -43,7 +48,7 @@ DreamWeaver 已经具备求职演示所需的完整闭环：
 
 最近一轮验证包括：
 
-- 后端全量测试：91 passed
+- 后端全量测试：94 passed
 - 后端 ruff：通过
 - 用户端生产构建：通过
 - 管理端生产构建：通过
@@ -56,10 +61,9 @@ DreamWeaver 已经具备求职演示所需的完整闭环：
 
 | Priority | Task | Why |
 | --- | --- | --- |
-| P0 | 将同步生成迁移到 Celery worker | 支持真实长任务、断点恢复和后台进度 |
+| P0 | 跨用户 / 跨环境 Provider dashboard | 当前已支持单用户摘要，后续要支持运营视角 |
-| P1 | 生成任务取消与重试队列 | 防止重复任务和用户误触造成浪费 |
-| P1 | 跨用户 / 跨环境 Provider dashboard | 当前已支持单用户摘要，后续要支持运营视角 |
 | P1 | 监控告警与结构化 dashboard | 目前已有故事库级概览，后续要接入更完整观测体系 |
+| P1 | 断点续跑与更细粒度任务控制 | 让取消、重试和 worker 恢复更稳 |
 | P2 | 更细粒度叙事风格与音色策略 | 扩展体验，但不影响当前求职版主线 |
 
 ---

docs/product/unified-generation-workflow-prd.md

@@ -68,21 +68,25 @@ DreamWeaver 当前同时支持普通故事生成、完整故事生成和绘本
 - 已新增任务运行概览 `GET /api/generations/ops-summary`，故事库可展示最近失败、运行中任务和超时待收敛任务
 - 重复资产任务已加入保护：同一故事存在运行中 job 时，不再重复触发封面、音频或统一资产重试
 - Celery beat 已支持定时收敛卡住的 generation job，避免任务长期停在 running
-- 用户端与管理端生成轨迹组件会在任务未终止时自动轮询，为后续后台 worker 进度流保留前端形态
+- 用户端与管理端生成轨迹组件会在任务未终止时自动轮询，已经可直接消费后台 worker 进度流
 - `POST /api/generations` 响应已返回 `generation_job_id`，smoke 脚本会验证 job 查询与 story job history
 - 用户端与管理端的故事详情页、绘本阅读页已接入生成轨迹，展示生成/重试任务、关键事件、Provider 调用结果和聚合指标
 - 故事详情页封面补全已切换到统一资产重试入口
 - 管理端前端构建阻塞已修复，主前端与 admin 前端均可完成生产构建
+- 已补首版生成任务控制能力：
+  - `POST /api/generations/jobs/{job_id}/cancel`
+  - `POST /api/generations/jobs/{job_id}/retry`
+  - 创建弹窗与生成轨迹都可触发取消或重新排队
 
 ### Remaining Production Work
 
 - 普通故事、完整生成、绘本生成已有统一外部入口，内部 workflow 仍可继续减少兼容层分支
 - 统一资产重试入口已覆盖普通故事封面、绘本缺失插图和故事音频，后续可继续扩展更细的资产级审计
-- 后台异步 worker 执行、断点续跑、跨用户/跨环境 Provider 分析，以及真正的取消/重试队列仍属于后续生产化增强
+- 断点续跑、跨用户/跨环境 Provider 分析，以及更细粒度的任务控制策略仍属于后续生产化增强
 
 ### What This Means
 
-这份 PRD 仍然保留目标态设计，但主干能力已经可在当前代码中演示。当前最适合的继续方式，是继续把同步请求迁移到后台 worker，并把当前首版运营摘要扩展为可筛选、可对比的分析视角，而不是继续扩大功能范围。
+这份 PRD 仍然保留目标态设计，但主干能力已经可在当前代码中演示。当前最适合的继续方式，是在已落地的 worker 化与任务控制基础上，把当前首版运营摘要扩展为可筛选、可对比的分析视角，并逐步补断点续跑和更完整监控，而不是继续扩大功能范围。
 
 ---
 
@@ -93,13 +97,13 @@ DreamWeaver 当前同时支持普通故事生成、完整故事生成和绘本
 DreamWeaver 当前存在以下工作流层面问题：
 
 1. **生成入口已建立，内部路径正在收束**
-   当前前端已切到 `/api/generations`，旧的 `/api/stories/generate`、`/api/stories/generate/full`、`/api/storybook/generate` 仍作为兼容入口保留。service 内部已抽取上下文准备、主记录保存、封面补全、绘本插图补全和音频补全 helper，并用 `AssetCompletionResult` 表达资产补全结果。generation job/event 已落库并可查询，Provider 调用轨迹、单故事聚合指标和跨故事运营摘要也已进入用户端与管理端展示。下一步重点是为后台异步 worker 复用这些事件。
+   当前前端已切到 `/api/generations`，旧的 `/api/stories/generate`、`/api/stories/generate/full`、`/api/storybook/generate` 仍作为兼容入口保留。service 内部已抽取上下文准备、主记录保存、封面补全、绘本插图补全和音频补全 helper，并用 `AssetCompletionResult` 表达资产补全结果。generation job/event 已落库并可查询，Provider 调用轨迹、单故事聚合指标和跨故事运营摘要也已进入用户端与管理端展示；统一生成请求现在已经交给后台 worker 执行。下一步重点是把取消/重试队列也接到这套事件模型上。
 
 2. **保存与资产补全过程正在统一**
    文本故事和绘本已拥有更清晰的主记录保存 helper；普通故事封面、绘本缺失插图、故事音频生成/缓存已共用各自的 asset completion helper。服务层已经能表达资产任务结果，并会把统一入口、资产重试、绘本逐页插图和音频生成的关键节点写入 job event。
 
 3. **状态表达已基本统一，仍需生产化扩展**
-   当前已经能用 `generation_status`、`text_status`、`image_status`、`audio_status` 和 `retryable_assets` 表达生成中、部分可读、完成、降级完成、失败和可重试。后续重点是让后台 worker、运营分析和通知系统复用同一套状态语义。
+   当前已经能用 `generation_status`、`text_status`、`image_status`、`audio_status` 和 `retryable_assets` 表达生成中、部分可读、完成、降级完成、失败和可重试。后续重点是让取消请求、重新排队、运营分析和通知系统复用同一套状态语义。
 
 4. **失败处理策略不统一**  
    图片、音频、绘本页生成失败时，系统没有统一的降级定义，用户体验和技术行为都不够稳定。

docs/technical/architecture.md

@@ -100,8 +100,7 @@ flowchart LR
 
 当前仍是求职版 MVP，不引入复杂工作流引擎。下一步生产化优先级：
 
-1. 把同步生成迁移到后台 worker。
+1. 补齐生成任务取消与重新排队能力，减少误触和重复消耗。
-2. 基于现有 job 查询和前端轮询展示真实异步进度。
+2. 基于现有 job 查询和前端轮询继续扩展真实异步进度与任务控制。
 3. 扩展 Provider analytics 的时间窗口、失败原因和跨用户维度。
-4. 为音频缓存增加过期策略和后台清理任务。
+4. 继续补充部署、监控告警和密钥治理策略。
-5. 补充部署、监控告警和密钥治理策略。

docs/technical/generation-job-state.md

@@ -6,7 +6,7 @@
 
 已新增轻量 `generation_jobs` 与 `generation_job_events` 表，但不引入复杂工作流引擎。
 
-原因是当前 MVP 的生成方式仍然以同步请求为主：后端在一次请求中完成主内容保存，再补全封面、绘本插图或语音。用户最关心的是“这个故事现在能不能读、哪些资产可补全”；系统侧则需要有足够的轨迹说明“这次生成做到了哪一步、哪里失败、哪些资产还能重试”。
+原因是当前 MVP 已经进入“请求接收与后台执行分离”的阶段：`POST /api/generations` 先接受请求并返回 `generation_job_id`，再由 Celery worker 完成主内容保存和后续资产补全。用户最关心的是“这个故事现在能不能读、任务跑到哪一步、哪些资产可补全”；系统侧则需要有足够的轨迹说明“这次生成做到了哪一步、哪里失败、是否被取消、哪些资产还能重试”。
 
 因此当前采用轻量落库策略：
 
@@ -43,22 +43,24 @@ job 响应会返回 `progress_percent`、`progress_label` 和 `is_terminal`，
 - 音频缓存由 `STORY_AUDIO_CACHE_TTL_DAYS` 控制过期时间，Celery beat 会每日清理。
 - 生成任务由 `GENERATION_JOB_STALE_MINUTES` 控制卡住阈值，Celery beat 会每 30 分钟扫描一次，将超时运行中的任务标记为 `generation_stale_failed`。
 - 当某个故事已经有运行中的 job 时，封面补全、音频生成和统一资产重试会直接拒绝重复请求，避免用户连点造成重复成本。
+- 统一生成请求已由 Celery worker 执行，前端会先拿到 `generation_job_id`，再轮询 job 详情直到主记录落库或任务终止。
+- 当前已支持首版任务控制：队列中的任务可直接取消，运行中的任务可在安全检查点取消，失败或已取消任务可重新排队。
 
-## 什么时候需要落库 job
+## 为什么当前仍然需要扩展 job 模型
 
-如果后续进入真实生产化，需要扩展当前 job/event 模型：
+虽然 worker 化已经完成，但如果继续进入真实生产化，仍然需要扩展当前 job/event 模型：
 
-- 生成流程改成真正异步，前端需要轮询后台 worker 的实时进度。
+- 单个故事会产生多次生成尝试，需要对比每次任务的 provider 表现、取消原因、重试原因和资产结果。
-- 单个故事会产生多次生成尝试，需要对比每次任务的 provider 表现、重试原因和资产结果。
 - 需要展示比当前事件更细颗粒度的步骤，例如 prompt 构建、provider 选择依据、provider failover 原因、每次调用 token/图片/语音成本。
 - 需要按 provider、成本、延迟和失败原因做运营分析。
+- 需要继续扩展取消与重试队列的颗粒度，例如更细的中断点、任务依赖和断点续跑策略。
 - 需要断点续跑，避免 Worker 重启后丢失中间状态。
 
 ## 推荐未来扩展
 
 当前已有两层记录，未来可以继续扩展字段和事件颗粒度：
 
-- 将同步生成请求迁移到真正异步 worker 后，继续复用现有 job 查询和前端轮询进度条。
+- 继续复用现有 job 查询和前端轮询进度条，为取消请求、重新排队和长任务通知提供统一入口。
 - 将当前跨故事 provider 指标扩展为跨用户、跨环境和更细颗粒度的失败原因维度分析。
 
 ## 面试表达

frontend/src/components/CreateStoryModal.vue

@@ -43,7 +43,10 @@ const outputMode = ref<'full_story' | 'storybook'>('full_story')
 const inputData = ref('')
 const educationTheme = ref('')
 const loading = ref(false)
+const canceling = ref(false)
+const cancelRequested = ref(false)
 const error = ref('')
+const activeGenerationJobId = ref<string | null>(null)
 
 // Data
 interface ChildProfile {
@@ -110,10 +113,17 @@ interface GenerationAcceptedResponse {
 
 interface GenerationJobDetail {
   story_id: number | null
+  status: string
+  current_step: string
   is_terminal: boolean
   error_message: string | null
 }
 
+interface GenerationJobActionResponse {
+  status: string
+  current_step: string
+}
+
 const JOB_POLL_INTERVAL_MS = 1500
 const JOB_POLL_MAX_ATTEMPTS = 80
 
@@ -121,6 +131,9 @@ const JOB_POLL_MAX_ATTEMPTS = 80
 function close() {
   emit('update:modelValue', false)
   error.value = ''
+  activeGenerationJobId.value = null
+  cancelRequested.value = false
+  canceling.value = false
 }
 
 function sleep(ms: number) {
@@ -132,6 +145,9 @@ function sleep(ms: number) {
 async function waitForStoryId(jobId: string) {
   for (let attempt = 0; attempt < JOB_POLL_MAX_ATTEMPTS; attempt += 1) {
     const detail = await api.get<GenerationJobDetail>(`/api/generations/jobs/${jobId}`)
+    if (detail.status === 'canceled' || detail.current_step === 'generation_canceled') {
+      return null
+    }
     if (detail.story_id) {
       return detail.story_id
     }
@@ -144,6 +160,27 @@ async function waitForStoryId(jobId: string) {
   throw new Error('任务已提交，但主内容落库超时，请稍后到故事库查看最新结果')
 }
 
+async function cancelGenerationJob() {
+  if (!activeGenerationJobId.value || canceling.value || cancelRequested.value) return
+
+  canceling.value = true
+  error.value = ''
+  try {
+    const result = await api.post<GenerationJobActionResponse>(
+      `/api/generations/jobs/${activeGenerationJobId.value}/cancel`,
+    )
+    cancelRequested.value = true
+    if (result.status === 'canceled' || result.current_step === 'generation_canceled') {
+      loading.value = false
+      close()
+    }
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '取消任务失败'
+  } finally {
+    canceling.value = false
+  }
+}
+
 async function fetchProfiles() {
   if (!userStore.user) return
   profileError.value = ''
@@ -192,6 +229,8 @@ async function generateStory() {
   }
   
   loading.value = true
+  cancelRequested.value = false
+  activeGenerationJobId.value = null
   error.value = ''
 
   try {
@@ -211,8 +250,13 @@ async function generateStory() {
     if (!jobId) {
       throw new Error('生成任务已创建，但缺少任务编号')
     }
+    activeGenerationJobId.value = jobId
 
     const storyId = accepted.id ?? await waitForStoryId(jobId)
+    if (storyId === null) {
+      close()
+      return
+    }
     close()
     if (requestedOutputMode.value === 'storybook') {
       router.push(`/storybook/view/${storyId}`)
@@ -223,6 +267,8 @@ async function generateStory() {
     error.value = e instanceof Error ? e.message : '生成失败'
   } finally {
     loading.value = false
+    activeGenerationJobId.value = null
+    cancelRequested.value = false
   }
 }
 </script>
@@ -253,6 +299,22 @@ async function generateStory() {
           :title="generationTitle"
           :steps="generationSteps"
         />
+        <div
+          v-if="loading && activeGenerationJobId"
+          class="fixed bottom-10 z-[110] flex flex-col items-center gap-3"
+        >
+          <BaseButton
+            variant="secondary"
+            :loading="canceling"
+            :disabled="cancelRequested"
+            @click="cancelGenerationJob"
+          >
+            {{ cancelRequested ? '正在取消任务...' : '取消任务' }}
+          </BaseButton>
+          <p class="text-sm text-white/70">
+            {{ cancelRequested ? '已提交取消请求，会在安全检查点停止任务。' : '如果是误触发起，可以现在取消后台任务。' }}
+          </p>
+        </div>
 
         <!-- 模态框内容 -->
         <div v-else class="relative w-full max-w-2xl max-h-[90vh] overflow-y-auto bg-[#1C2035] border border-gray-700/50 rounded-3xl shadow-2xl p-6 md:p-8">

frontend/src/components/GenerationTrace.vue

@@ -28,6 +28,7 @@ const jobHistory = ref<GenerationJobSummary[]>([])
 const activeJob = ref<GenerationJobDetail | null>(null)
 const providerStats = ref<GenerationProviderStats | null>(null)
 const loading = ref(false)
+const actionLoading = ref(false)
 const error = ref('')
 let refreshTimer: ReturnType<typeof setInterval> | null = null
 
@@ -79,6 +80,7 @@ const jobStatusClassMap: Record<string, string> = {
   succeeded: 'border-emerald-200 bg-emerald-50 text-emerald-700',
   completed: 'border-emerald-200 bg-emerald-50 text-emerald-700',
   degraded_completed: 'border-orange-200 bg-orange-50 text-orange-700',
+  canceled: 'border-slate-200 bg-slate-100 text-slate-700',
   failed: 'border-rose-200 bg-rose-50 text-rose-700',
 }
 
@@ -93,6 +95,7 @@ function getJobStatusLabel(status?: string) {
     succeeded: '成功',
     completed: '已完成',
     degraded_completed: '降级完成',
+    canceled: '已取消',
     failed: '失败',
   }
   return labels[status ?? ''] ?? '未知'
@@ -102,6 +105,8 @@ function getEventLabel(eventType: string) {
   const labels: Record<string, string> = {
     request_accepted: '请求接收',
     worker_started: '后台任务开始',
+    retry_queued: '重新排队',
+    cancel_requested: '已请求取消',
     context_prepared: '上下文准备',
     narrative_generated: '正文生成',
     story_saved: '故事保存',
@@ -124,6 +129,7 @@ function getEventLabel(eventType: string) {
     asset_retry_started: '资源重试开始',
     asset_retry_completed: '资源重试完成',
     asset_retry_failed: '资源重试失败',
+    generation_canceled: '任务已取消',
     generation_completed: '生成完成',
     generation_failed: '生成失败',
     asset_generation_completed: '资源生成完成',
@@ -202,6 +208,36 @@ async function refresh() {
   }
 }
 
+async function cancelActiveJob() {
+  if (!activeJob.value || actionLoading.value) return
+
+  actionLoading.value = true
+  error.value = ''
+  try {
+    await api.post(`/api/generations/jobs/${activeJob.value.id}/cancel`)
+    await refresh()
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '取消任务失败'
+  } finally {
+    actionLoading.value = false
+  }
+}
+
+async function retryActiveJob() {
+  if (!activeJob.value || actionLoading.value) return
+
+  actionLoading.value = true
+  error.value = ''
+  try {
+    await api.post(`/api/generations/jobs/${activeJob.value.id}/retry`)
+    await refresh()
+  } catch (e) {
+    error.value = e instanceof Error ? e.message : '重新排队失败'
+  } finally {
+    actionLoading.value = false
+  }
+}
+
 function stopAutoRefresh() {
   if (refreshTimer) {
     clearInterval(refreshTimer)
@@ -323,10 +359,30 @@ defineExpose({ refresh })
                 当前步骤：{{ getEventLabel(activeJob.current_step) }}
               </div>
             </div>
+            <div class="flex flex-wrap items-center justify-end gap-2">
+              <button
+                v-if="activeJob.can_cancel"
+                type="button"
+                class="rounded-full border border-amber-200 bg-amber-50 px-3 py-1 text-xs font-medium text-amber-700 transition hover:bg-amber-100 disabled:cursor-not-allowed disabled:opacity-60"
+                :disabled="actionLoading"
+                @click="cancelActiveJob"
+              >
+                {{ actionLoading ? '处理中...' : '取消任务' }}
+              </button>
+              <button
+                v-if="activeJob.can_retry"
+                type="button"
+                class="rounded-full border border-sky-200 bg-sky-50 px-3 py-1 text-xs font-medium text-sky-700 transition hover:bg-sky-100 disabled:cursor-not-allowed disabled:opacity-60"
+                :disabled="actionLoading"
+                @click="retryActiveJob"
+              >
+                {{ actionLoading ? '处理中...' : '重新排队' }}
+              </button>
               <span class="rounded-full border px-3 py-1 text-xs font-medium" :class="getJobStatusClass(activeJob.status)">
                 {{ getJobStatusLabel(activeJob.status) }}
               </span>
             </div>
+          </div>
 
           <div>
             <div class="mb-1 flex items-center justify-between text-xs" :class="mutedTextClass">

frontend/src/types/generation.ts

@@ -8,6 +8,8 @@ export interface GenerationJobSummary {
   progress_percent: number
   progress_label: string
   is_terminal: boolean
+  can_cancel: boolean
+  can_retry: boolean
   result_snapshot: Record<string, unknown>
   error_message: string | null
   created_at: string