feat(extract): 抽帧打印进度日志

ffmpeg 增加 -progress pipe:1 并解析 frame=N：每约 10 个检查点打印
'抽帧进度 X/Y 帧 (Z 帧/s)'，长视频抽帧期间可见实时进度与处理速度。
测试覆盖进度行解析、进度日志输出与失败路径（Popen mock）。

nodes/frame_extract.py

@@ -12,6 +12,7 @@ from __future__ import annotations
 import json
 import re
 import subprocess
+import time
 from pathlib import Path
 
 from nodes.ffmpeg import _ffmpeg_bin
@@ -96,6 +97,20 @@ def _frame_step(fps: float, interval: float) -> int:
     return max(1, int(round(interval * fps)))
 
 
+def _parse_progress_line(line: str) -> int | None:
+    """解析 ffmpeg -progress 输出行，返回 frame=N 的数值；其他行返回 None。
+
+    -progress 每 ~0.5s 输出一组 key=value（frame/fps/out_time_ms/progress 等），
+    这里只关心 frame=（已写出的选中帧数），用于实时进度与速度统计。
+    """
+    stripped = line.strip()
+    if not stripped.startswith("frame="):
+        return None
+    try:
+        return int(stripped.split("=", 1)[1])
+    except ValueError:
+        return None
+
 def invoke(request: InvokeRequest) -> InvokeResponse:
     """按帧间隔抽取并裁切视频帧，输出 frames.json 清单。"""
     video_uri = request.inputs.get("video_uri")
@@ -137,13 +152,18 @@ def invoke(request: InvokeRequest) -> InvokeResponse:
     # 单次解码全片：select 按帧号（n mod step == 0）精确取帧，随后对选中帧
     # 裁切字幕区域并压缩到 720p 内（仅缩小）——过大的输入图会触发 glm-ocr
     # 的重复循环 bug（M-RoPE delta），源头规避。
-    result = subprocess.run(
+    # -progress 把 frame=N 进度写入管道，实时打印抽帧进度与速度。
+    expected_frames = (int(duration * fps) + step - 1) // step
+    process = subprocess.Popen(
         [
             ffmpeg_bin,
             "-y",
             "-hide_banner",
             "-loglevel",
             "error",
+            "-nostats",
+            "-progress",
+            "pipe:1",
             "-i",
             str(video),
             "-vf",
@@ -157,11 +177,30 @@ def invoke(request: InvokeRequest) -> InvokeResponse:
             "vfr",
             str(frames_dir / "frame_%04d.png"),
         ],
-        capture_output=True,
+        stdout=subprocess.PIPE,
+        stderr=subprocess.PIPE,
         text=True,
     )
-    if result.returncode != 0:
-        return InvokeResponse(status="failed", error=result.stderr[-500:] or "ffmpeg failed")
+    extract_started = time.monotonic()
+    last_logged = 0
+    progress_step = max(1, expected_frames // 10)
+    for line in process.stdout:
+        done = _parse_progress_line(line)
+        if done is None:
+            continue
+        # 每写出约 10 个检查点打一次日志，避免刷屏；同时给出处理速度。
+        if done - last_logged >= progress_step:
+            elapsed = time.monotonic() - extract_started
+            logger.info(
+                "抽帧进度 %d/%d 帧 (%.1f 帧/s)",
+                min(done, expected_frames), expected_frames,
+                done / elapsed if elapsed > 0 else 0.0,
+            )
+            last_logged = done
+    process.wait()
+    stderr = process.stderr.read()
+    if process.returncode != 0:
+        return InvokeResponse(status="failed", error=stderr[-500:] or "ffmpeg failed")
 
     # 第 k 个输出文件对应原始帧号 k×step，时间 = 帧号 / fps（帧精确，无累计偏差）。
     files = sorted(frames_dir.glob("frame_*.png"))

tests/test_ocr_flow.py

@@ -159,6 +159,84 @@ def test_video_duration_unreadable(monkeypatch) -> None:
     assert _video_duration(Path("/tmp/x.mp4"), "ffmpeg") is None
 
 
+class _FakeReader:
+    """模拟 stderr 读取对象。"""
+
+    def __init__(self, content: str = "") -> None:
+        self._content = content
+
+    def read(self) -> str:
+        return self._content
+
+
+class FakePopen:
+    """模拟 ffmpeg 进程：stdout 可迭代 -progress 行，可配置返回码与 stderr。"""
+
+    def __init__(self, lines=(), returncode: int = 0, stderr: str = "") -> None:
+        self.stdout = list(lines)
+        self.stderr = _FakeReader(stderr)
+        self.returncode = returncode
+
+    def wait(self) -> int:
+        return self.returncode
+
+
+def _fake_popen(lines=(), returncode: int = 0, stderr: str = ""):
+    """构造替换 subprocess.Popen 的工厂函数。"""
+    return lambda *a, **k: FakePopen(lines, returncode, stderr)
+
+
+def test_parse_progress_line() -> None:
+    """-progress 行解析：frame=N 返回数值，其他行与非法值返回 None。"""
+    from nodes.frame_extract import _parse_progress_line
+
+    assert _parse_progress_line("frame=25\n") == 25
+    assert _parse_progress_line("progress=continue") is None
+    assert _parse_progress_line("fps=25.0") is None
+    assert _parse_progress_line("frame=abc") is None
+
+
+def test_frame_extract_progress_logging(monkeypatch, tmp_path) -> None:
+    """ffmpeg -progress 的 frame=N 被解析并打印抽帧进度与速度。"""
+    import logging
+
+    monkeypatch.setattr("nodes.frame_extract._video_size", lambda *a, **k: (1280, 720))
+    monkeypatch.setattr("nodes.frame_extract._video_fps", lambda *a, **k: 25.0)
+    monkeypatch.setattr("nodes.frame_extract._video_duration", lambda *a, **k: 10.0)
+    monkeypatch.setattr(
+        "nodes.frame_extract.subprocess.Popen",
+        _fake_popen(lines=["fps=25.0", "frame=20", "progress=continue", "frame=50", "progress=continue"]),
+    )
+    captured: list[str] = []
+
+    class CaptureHandler(logging.Handler):
+        def emit(self, record):
+            captured.append(record.getMessage())
+
+    logger = logging.getLogger("vrsub.frame-extract")
+    logger.addHandler(CaptureHandler())
+    try:
+        response = frame_invoke(_frame_request(tmp_path))
+    finally:
+        logger.removeHandler(logger.handlers[-1])
+    assert response.status == "completed", response.error
+    assert any("抽帧进度" in message and "帧/s" in message for message in captured)
+
+
+def test_frame_extract_ffmpeg_fails(monkeypatch, tmp_path) -> None:
+    """ffmpeg 抽帧失败时透传错误。"""
+    monkeypatch.setattr("nodes.frame_extract._video_size", lambda *a, **k: (1280, 720))
+    monkeypatch.setattr("nodes.frame_extract._video_fps", lambda *a, **k: 25.0)
+    monkeypatch.setattr("nodes.frame_extract._video_duration", lambda *a, **k: 10.0)
+    monkeypatch.setattr(
+        "nodes.frame_extract.subprocess.Popen",
+        _fake_popen(lines=["frame=1", "progress=end"], returncode=1, stderr="boom"),
+    )
+    response = frame_invoke(_frame_request(tmp_path))
+    assert response.status == "failed"
+    assert "boom" in response.error
+
+
 def test_frame_extract_duration_unknown(monkeypatch, tmp_path) -> None:
     """无法读取视频时长时返回失败。"""
     monkeypatch.setattr("nodes.frame_extract._video_size", lambda *a, **k: (1280, 720))