auto-save 2026-05-14 04:32 (~5)

2026-05-14 04:32:27 +08:00
parent 8f2b8d373c
commit 4935e34eb0
5 changed files with 106 additions and 21 deletions
--- a/.memory/worklog.json
+++ b/.memory/worklog.json
@@ -3199,6 +3199,19 @@
      "type": "session-heartbeat",
      "message": "Claude 会话活跃 · 最近命令：claude · 4 项未提交变更 · 最近提交：auto-save 2026-05-14 04:21 (~6)",
      "files_changed": 4
+    },
+    {
+      "ts": "2026-05-14T04:26:56+08:00",
+      "type": "commit",
+      "message": "auto-save 2026-05-14 04:26 (~5)",
+      "hash": "8f2b8d3",
+      "files_changed": 5
+    },
+    {
+      "ts": "2026-05-13T20:28:50Z",
+      "type": "session-heartbeat",
+      "message": "Codex 会话活跃 · 最近命令：codex · 1 项未提交变更 · 最近提交：auto-save 2026-05-14 04:26 (~5)",
+      "files_changed": 1
    }
  ]
 }
--- a/api/main.py
+++ b/api/main.py
@@ -90,7 +90,8 @@ JobStatus = Literal[
 KEYFRAME_COUNT = int(os.getenv("KEYFRAME_COUNT", "5"))
 FrameExtractTarget = Literal["balanced", "subject", "transition", "expression", "motion"]
 FrameExtractMode = Literal["replace", "append"]
-FrameExtractQuality = Literal["fast", "accurate", "ultra"]
+FrameExtractQuality = Literal["auto", "fast", "accurate", "ultra"]
+AnalyzeTask = tuple[str, int, FrameExtractTarget, FrameExtractMode, FrameExtractQuality]
 FRAME_TARGET_LABELS: dict[FrameExtractTarget, str] = {
    "balanced": "综合关键帧",
    "subject": "清晰主体",
@@ -99,6 +100,7 @@ FRAME_TARGET_LABELS: dict[FrameExtractTarget, str] = {
    "motion": "动作峰值",
 }
 FRAME_QUALITY_LABELS: dict[FrameExtractQuality, str] = {
+    "auto": "自动",
    "fast": "快速",
    "accurate": "精细",
    "ultra": "极准",
@@ -221,6 +223,8 @@ class Job(BaseModel):


 JOBS: dict[str, Job] = {}
+ANALYZE_QUEUE: list[AnalyzeTask] = []
+ANALYZE_WORKER_RUNNING = False


 def job_dir(job_id: str) -> Path:
@@ -441,6 +445,30 @@ def _frame_metrics(img_path: Path, idx: int, timestamp: float, metric_width: int
    }


+def _physical_memory_gb() -> float:
+    try:
+        page_size = os.sysconf("SC_PAGE_SIZE")
+        pages = os.sysconf("SC_PHYS_PAGES")
+        return float(page_size * pages) / (1024 ** 3)
+    except Exception:
+        return 0.0
+
+
+def _resolve_frame_quality(duration: float, quality: FrameExtractQuality) -> FrameExtractQuality:
+    if quality != "auto":
+        return quality
+    cores = os.cpu_count() or 4
+    memory_gb = _physical_memory_gb()
+    strong_machine = cores >= 10 and (memory_gb == 0.0 or memory_gb >= 32)
+    if strong_machine and duration <= 180:
+        return "ultra"
+    if strong_machine and duration <= 600:
+        return "accurate"
+    if cores >= 8 and duration <= 240:
+        return "accurate"
+    return "fast"
+
+
 def _scan_profile(duration: float, quality: FrameExtractQuality) -> tuple[float, int, int, int]:
    """返回 scan_fps / scan_width / metric_width / estimated_count。"""
    if quality == "ultra":
@@ -607,7 +635,7 @@ async def pipeline_analyze(
    frame_count: int = KEYFRAME_COUNT,
    target: FrameExtractTarget = "balanced",
    mode: FrameExtractMode = "replace",
-    quality: FrameExtractQuality = "accurate",
+    quality: FrameExtractQuality = "auto",
 ) -> None:
    """阶段 2：拆音轨 + 抽关键帧。ASR/翻译是独立文案轨，不阻塞视觉素材流。"""
    job = JOBS[job_id]
@@ -630,9 +658,11 @@ async def pipeline_analyze(

        n = max(1, min(int(frame_count), 20))
        target_label = FRAME_TARGET_LABELS.get(target, FRAME_TARGET_LABELS["balanced"])
-        quality_label = FRAME_QUALITY_LABELS.get(quality, FRAME_QUALITY_LABELS["accurate"])
        duration = max(float(job.duration or 1.0), 0.1)
-        scan_fps, scan_width, metric_width, estimated_scan_count = _scan_profile(duration, quality)
+        effective_quality = _resolve_frame_quality(duration, quality)
+        effective_quality_label = FRAME_QUALITY_LABELS.get(effective_quality, FRAME_QUALITY_LABELS["accurate"])
+        quality_label = f"自动·{effective_quality_label}" if quality == "auto" else effective_quality_label
+        scan_fps, scan_width, metric_width, estimated_scan_count = _scan_profile(duration, effective_quality)

        update(job, message=f"本地{quality_label}扫描 · {target_label} · 约 {estimated_scan_count} 帧…", progress=45)
        frames_dir = d / "frames"
@@ -716,6 +746,24 @@ async def pipeline_analyze(
        update(job, status="failed", error=str(e), message="解析失败")


+async def analyze_queue_worker() -> None:
+    global ANALYZE_WORKER_RUNNING
+    ANALYZE_WORKER_RUNNING = True
+    try:
+        while ANALYZE_QUEUE:
+            job_id, frames, target, mode, quality = ANALYZE_QUEUE.pop(0)
+            if job_id not in JOBS:
+                continue
+            await pipeline_analyze(job_id, frames, target, mode, quality)
+            if ANALYZE_QUEUE:
+                for pos, (queued_job_id, *_rest) in enumerate(ANALYZE_QUEUE, start=1):
+                    queued_job = JOBS.get(queued_job_id)
+                    if queued_job:
+                        update(queued_job, status="splitting", progress=30, message=f"排队等待抽帧 · 前方 {pos - 1} 个任务")
+    finally:
+        ANALYZE_WORKER_RUNNING = False
+
+
 # ---------- Gemini ASR + 翻译 ----------

 def _transcribe_sync(wav: Path) -> list[dict]:
@@ -1084,14 +1132,23 @@ async def trigger_analyze(
    frames: int = KEYFRAME_COUNT,
    target: FrameExtractTarget = "balanced",
    mode: FrameExtractMode = "replace",
-    quality: FrameExtractQuality = "accurate",
+    quality: FrameExtractQuality = "auto",
 ) -> Job:
    job = JOBS.get(job_id)
    if not job:
        raise HTTPException(404, "job not found")
    if job.status not in {"downloaded", "frames_extracted", "transcribed", "failed"}:
        raise HTTPException(409, f"status must be downloaded/failed, got {job.status}")
-    bg.add_task(pipeline_analyze, job_id, frames, target, mode, quality)
+    ANALYZE_QUEUE.append((job_id, frames, target, mode, quality))
+    position = len(ANALYZE_QUEUE)
+    update(
+        job,
+        status="splitting",
+        progress=30,
+        message="排队等待抽帧" if ANALYZE_WORKER_RUNNING or position > 1 else "准备抽帧…",
+    )
+    if not ANALYZE_WORKER_RUNNING:
+        bg.add_task(analyze_queue_worker)
    return job


--- a/web/app/page.tsx
+++ b/web/app/page.tsx
@@ -42,6 +42,7 @@ const FRAME_TARGET_LABELS: Record<FrameExtractTarget, string> = {
  motion: "动作峰值",
 }
 const FRAME_QUALITY_LABELS: Record<FrameExtractQuality, string> = {
+  auto: "自动",
  fast: "快速",
  accurate: "精细",
  ultra: "极准",
@@ -178,7 +179,7 @@ export default function Home() {
    if (!targetJob) return
    const frameTarget = frameTargets[jobId] ?? "balanced"
    const frameCount = frameCounts[jobId] ?? 5
-    const frameQuality = frameQualities[jobId] ?? "ultra"
+    const frameQuality = frameQualities[jobId] ?? "auto"
    const mode = options?.mode ?? (targetJob.frames.length > 0 ? "append" : "replace")
    setActiveJobId(jobId)
    setAnalyzing(true)
@@ -497,30 +498,43 @@ export default function Home() {
    })
  }, [job?.id, job?.frames])

-  // 轮询 Job（downloaded / transcribed / failed 三态停止）
+  // 轮询 Job：任一视频在下载 / 抽帧 / 生视频时都继续轮询，支持多个抽帧任务排队。
  const prevStatusRef = useRef<string | null>(null)
  useEffect(() => {
-    if (!job) return
+    if (jobs.length === 0) return
    // 状态切到 downloaded 时提示用户点解析（仅一次）
-    if (job.status === "downloaded" && prevStatusRef.current !== "downloaded") {
+    if (job?.status === "downloaded" && prevStatusRef.current !== "downloaded") {
      toast.info("📥 视频已就绪 — 请点 Input 节点里的「点这里开始解析」按钮", { duration: 6000 })
    }
-    prevStatusRef.current = job.status
+    prevStatusRef.current = job?.status ?? null

-    const runningVideo = !!job.generated_videos?.some((v) => v.status === "queued" || v.status === "in_progress")
    const TERMINAL: Job["status"][] = ["downloaded", "frames_extracted", "transcribed", "failed"]
-    if (TERMINAL.includes(job.status) && !runningVideo) {
+    const runningIds = jobs
+      .filter((item) => {
+        const runningVideo = !!item.generated_videos?.some((v) => v.status === "queued" || v.status === "in_progress")
+        return runningVideo || !TERMINAL.includes(item.status)
+      })
+      .map((item) => item.id)
+
+    if (runningIds.length === 0) {
      if (pollRef.current) { clearInterval(pollRef.current); pollRef.current = null }
      return
    }
    pollRef.current = setInterval(async () => {
      try {
-        const latest = await getJob(job.id)
-        setJob(latest)
+        const latestJobs = await Promise.all(runningIds.map((id) => getJob(id).catch(() => null)))
+        const byId = new Map(latestJobs.filter((item): item is Job => !!item).map((item) => [item.id, item]))
+        if (byId.size > 0) {
+          setJobs((prev) => prev.map((item) => byId.get(item.id) ?? item))
+        }
      } catch { /* silent */ }
    }, 1500)
    return () => { if (pollRef.current) clearInterval(pollRef.current) }
-  }, [job?.id, job?.status, job?.generated_videos?.map((v) => `${v.id}:${v.status}:${v.progress}`).join("|")])
+  }, [
+    job?.id,
+    job?.status,
+    jobs.map((item) => `${item.id}:${item.status}:${item.progress}:${item.generated_videos?.map((v) => `${v.id}:${v.status}:${v.progress}`).join(",")}`).join("|"),
+  ])

  const [pinnedNodes, setPinnedNodes] = useState<Set<string>>(() => new Set(loadNodePins()))
  const handleToggleNodePin = useCallback((id: string) => {
--- a/web/components/nodes/index.tsx
+++ b/web/components/nodes/index.tsx
@@ -135,6 +135,7 @@ const FRAME_TARGET_OPTIONS: Array<{ value: FrameExtractTarget; label: string; hi
 ]
 const FRAME_COUNT_OPTIONS = [3, 5, 8, 12]
 const FRAME_QUALITY_OPTIONS: Array<{ value: FrameExtractQuality; label: string; hint: string }> = [
+  { value: "auto", label: "自动", hint: "按电脑性能和视频时长自动选择" },
  { value: "fast", label: "快速", hint: "2fps / 360px，长视频省电" },
  { value: "accurate", label: "精细", hint: "8fps / 720px，M2 Max 轻松可用" },
  { value: "ultra", label: "极准", hint: "12fps / 960px，本机约 3 秒扫描 1 分钟视频" },
@@ -438,7 +439,7 @@ function FrameExtractQuickBar({
  onAnalyze: () => void
 }) {
  const option = FRAME_TARGET_OPTIONS.find((item) => item.value === target) ?? FRAME_TARGET_OPTIONS[0]
-  const qualityOption = FRAME_QUALITY_OPTIONS.find((item) => item.value === quality) ?? FRAME_QUALITY_OPTIONS[1]
+  const qualityOption = FRAME_QUALITY_OPTIONS.find((item) => item.value === quality) ?? FRAME_QUALITY_OPTIONS[0]
  const [settingsOpen, setSettingsOpen] = useState(false)

  return (
@@ -569,7 +570,7 @@ export function InputNode({ data, selected }: NodeProps<{ data: NodeData }> | an
            const toolWidth = Math.max(148, thumbNaturalWidth)
            const target = d.frameTargets[j.id] ?? "balanced"
            const count = d.frameCounts[j.id] ?? 5
-            const quality = d.frameQualities[j.id] ?? "ultra"
+            const quality = d.frameQualities[j.id] ?? "auto"
            const jHasFrames = j.frames.length > 0
            const jRunning = ["splitting", "transcribing"].includes(j.status)
            return (
@@ -583,7 +584,7 @@ export function InputNode({ data, selected }: NodeProps<{ data: NodeData }> | an
                    target={target}
                    count={count}
                    quality={quality}
-                    disabled={jRunning || d.analyzing}
+                    disabled={jRunning}
                    running={jRunning}
                    hasFrames={jHasFrames}
                    onTargetChange={(next) => d.onFrameTargetChange(j.id, next)}
--- a/web/lib/api.ts
+++ b/web/lib/api.ts
@@ -130,7 +130,7 @@ export interface KeyFrame {

 export type FrameExtractTarget = "balanced" | "subject" | "transition" | "expression" | "motion"
 export type FrameExtractMode = "replace" | "append"
-export type FrameExtractQuality = "fast" | "accurate" | "ultra"
+export type FrameExtractQuality = "auto" | "fast" | "accurate" | "ultra"

 export interface TranscriptSegment {
  index: number
@@ -268,7 +268,7 @@ export async function analyzeJob(
  frames = 5,
  target: FrameExtractTarget = "balanced",
  mode: FrameExtractMode = "replace",
-  quality: FrameExtractQuality = "accurate",
+  quality: FrameExtractQuality = "auto",
 ): Promise<Job> {
  const qs = new URLSearchParams({ frames: String(frames), target, mode, quality })
  const res = await fetch(`${API_BASE}/jobs/${id}/analyze?${qs.toString()}`, { method: "POST" })