import shutil import subprocess import tempfile from datetime import datetime, timezone from pathlib import Path from celery.utils.log import get_task_logger from video_processing.oss_helpers import download_asset from worker_app.celery_app import celery_app from worker_app.core.asset_types import infer_mime_type_from_storage_key from worker_app.db import SessionLocal from packages.adapters.sqlalchemy_impl import ( SQLAlchemyAssetRepository, SQLAlchemyIngestJobRepository, ) from packages.domain import Asset, AssetStatus, IngestJobStatus from packages.domain.media_validation import is_valid_media as _is_valid_media from packages.domain.media_validation import safe_parse_fps as _safe_parse_fps logger = get_task_logger(__name__) def extract_media_metadata(file_url: str, media_type: str) -> tuple[dict, bool]: """ 提取媒体文件的元数据。 Args: file_url: 媒体文件 URL 或本地路径 media_type: 媒体类型 (video, audio, image) Returns: (metadata_dict, success) - metadata_dict: 提取的元数据字典，失败时返回空字典 - success: 是否成功提取到有效元数据 """ metadata = {} success = False try: if media_type == "video": # 使用 ffprobe 提取视频元数据 from video_processing.ffmpeg_utils import run_ffprobe cmd = [ "ffprobe", "-v", "quiet", "-print_format", "json", "-show_format", "-show_streams", file_url, ] try: stdout, _ = run_ffprobe(cmd, timeout=30) import json as json_lib probe_data = json_lib.loads(stdout) # 提取视频流信息 for stream in probe_data.get("streams", []): if stream.get("codec_type") == "video": metadata["width"] = int(stream.get("width", 0)) metadata["height"] = int(stream.get("height", 0)) metadata["codec"] = stream.get("codec_name", "") metadata["fps"] = ( _safe_parse_fps(stream.get("r_frame_rate", "0/1")) if stream.get("r_frame_rate") else 0 # type: ignore[assignment] ) break # 提取格式信息 format_info = probe_data.get("format", {}) metadata["duration"] = float(format_info.get("duration", 0)) # type: ignore[assignment] metadata["size_bytes"] = int(format_info.get("size", 0)) metadata["bitrate"] = int(format_info.get("bit_rate", 0)) # 有效性判断：有视频流 + duration > 0 + size > 0 has_video_stream = any(s.get("codec_type") == "video" for s in probe_data.get("streams", [])) has_audio_stream = any(s.get("codec_type") == "audio" for s in probe_data.get("streams", [])) if (has_video_stream or has_audio_stream) and metadata.get("duration", 0) > 0: success = True except Exception as e: logger.warning("视频元数据提取失败: %s", e) elif media_type == "audio": from video_processing.ffmpeg_utils import run_ffprobe cmd = [ "ffprobe", "-v", "quiet", "-print_format", "json", "-show_format", "-show_streams", file_url, ] try: stdout, _ = run_ffprobe(cmd, timeout=30) import json as json_lib probe_data = json_lib.loads(stdout) has_audio_stream = any(s.get("codec_type") == "audio" for s in probe_data.get("streams", [])) format_info = probe_data.get("format", {}) metadata["duration"] = float(format_info.get("duration", 0)) # type: ignore[assignment] metadata["size_bytes"] = int(format_info.get("size", 0)) metadata["bitrate"] = int(format_info.get("bit_rate", 0)) metadata["codec"] = next( (s.get("codec_name", "") for s in probe_data.get("streams", []) if s.get("codec_type") == "audio"), "", ) if has_audio_stream and metadata.get("duration", 0) > 0: success = True except Exception as e: logger.warning("音频元数据提取失败: %s", e) elif media_type == "image": # 使用 Pillow 提取图片元数据 try: from PIL import Image with Image.open(file_url) as img: img.verify() # 验证文件完整性 # verify后需要重新打开才能读尺寸 with Image.open(file_url) as img2: metadata["width"] = img2.width metadata["height"] = img2.height metadata["format"] = img2.format metadata["mode"] = img2.mode if img2.width > 0 and img2.height > 0: success = True except ImportError: logger.warning("Pillow not available for image metadata extraction") except Exception as e: logger.warning(f"Failed to extract image metadata: {e}") except subprocess.TimeoutExpired: logger.warning(f"Timeout extracting metadata from {file_url}") except FileNotFoundError: logger.warning("ffprobe not found, cannot extract video metadata") except Exception as e: logger.warning(f"Failed to extract metadata: {e}") return metadata, success @celery_app.task(name="worker.ingest_asset") def ingest_asset(job_id: str) -> dict: """ Ingest asset task. Steps: 1. Fetch IngestJob from repository 2. Extract metadata from storage_key 3. Create Asset entity 4. Update IngestJob status to COMPLETED 5. Return result """ db = SessionLocal() try: job_repo = SQLAlchemyIngestJobRepository(db) asset_repo = SQLAlchemyAssetRepository(db) job = job_repo.get(job_id) if job is None: return {"status": "failed", "error": "job not found"} # Update job status to PROCESSING job.status = IngestJobStatus.PROCESSING job.updated_at = datetime.now(timezone.utc) job_repo.update(job) db.commit() # Extract real metadata from media file filename = job.storage_key.split("/")[-1] mime_type = infer_mime_type_from_storage_key(job.storage_key) # Determine media type from mime_type media_type = "video" if mime_type.startswith("image/"): media_type = "image" elif mime_type.startswith("audio/"): media_type = "audio" # 先从 OSS 下载文件到本地临时目录，再提取元数据 # （storage_key 是 OSS 内部路径，不能直接传给 ffprobe/Pillow） local_file = None thumbnail_url = None try: suffix = Path(job.storage_key).suffix or ".bin" with tempfile.NamedTemporaryFile(suffix=suffix, delete=False) as tmp: local_file = Path(tmp.name) download_ok = download_asset(job.storage_key, local_file) if not download_ok: logger.warning("素材下载失败，无法提取元数据: job_id=%s storage_key=%s", job_id, job.storage_key) metadata, extract_success = {}, False else: metadata, extract_success = extract_media_metadata(str(local_file), media_type) # 视频类型：生成缩略图（文件还在的时候生成） thumbnail_url = None if media_type == "video" and extract_success: frame_path = None try: from video_processing.oss_helpers import upload_to_oss from video_processing.thumbnail_generator import extract_first_frame frame_path = extract_first_frame(str(local_file), width=640) thumb_storage_key = f"assets/{job.project_id}/thumbnails/{job_id}.jpg" try: thumbnail_url = upload_to_oss(frame_path, thumb_storage_key) finally: if frame_path: try: Path(frame_path).unlink(missing_ok=True) except Exception: pass if thumbnail_url: logger.info( "素材缩略图生成成功: job_id=%s url=%s", job_id, thumbnail_url[:80], ) except Exception as thumb_err: logger.warning( "素材缩略图生成失败（不影响主流程）: job_id=%s error=%s", job_id, thumb_err, ) # ── HEVC 自动转码为 1080p H.264 ────────────────────────────── # 浏览器 WebCodecs 硬件解码 HEVC 输出黑帧，上传时自动转码 # 失败时降级使用原始文件，不阻塞上传流程 if media_type == "video" and local_file and local_file.exists(): codec = (metadata.get("codec") or "").lower() if codec in ("hevc", "h265", "hvh1"): logger.info( "检测到 HEVC 编码 (codec=%s)，启动转码: job_id=%s", codec, job_id, ) _tc_tmp = None _needs_rotation = False # ── Step 1: 磁盘空间检查（独立 try/except，失败仍尝试转码）── try: _disk_usage = shutil.disk_usage("/tmp") _free_gb = _disk_usage.free / (1024**3) if _free_gb < 2: raise RuntimeError(f"磁盘空间不足 ({_free_gb:.1f}GB < 2GB)") except Exception as _disk_err: logger.warning("磁盘检查失败，仍尝试转码: job_id=%s err=%s", job_id, _disk_err) # ── Step 2: ffprobe 旋转检测（独立 try/except，失败不阻塞转码）── try: _probe_cmd = [ "ffprobe", "-v", "error", "-select_streams", "v:0", "-show_entries", "side_data=rotation", "-show_entries", "stream_tags=rotate", "-of", "default=noprint_wrappers=1:nokey=1", str(local_file), ] _probe_result = subprocess.run( _probe_cmd, stdout=subprocess.PIPE, stderr=subprocess.DEVNULL, text=True, timeout=60, # 大文件在容器 overlay 文件系统上解析可能较慢 ) _rotation_str = (_probe_result.stdout or "").strip().split("\n")[0] if _rotation_str in ("90", "270", "-90"): _needs_rotation = True logger.info( "检测到竖屏视频 (rotation=%s)，将物理旋转画面: job_id=%s", _rotation_str, job_id, ) except subprocess.TimeoutExpired: logger.warning( "ffprobe 旋转检测超时(60s)，跳过旋转继续转码: job_id=%s", job_id, ) _needs_rotation = False except Exception as _probe_err: logger.warning( "ffprobe 旋转检测异常，跳过旋转继续转码: job_id=%s err=%s", job_id, _probe_err, ) _needs_rotation = False # ── Step 3: ffmpeg 转码（独立 try/except）── try: _tc_tmp_file = tempfile.NamedTemporaryFile(delete=False, suffix="_h264.mp4") _tc_tmp = Path(_tc_tmp_file.name) _tc_tmp_file.close() # 关闭文件描述符，ffmpeg 会自己打开 # 构建 video filter：竖屏先旋转再缩放 if _needs_rotation: _vf = "transpose=1,scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'" else: _vf = "scale='if(gt(ih,1080),-2,iw)':'if(gt(ih,1080),1080,ih)'" _cmd = [ "ffmpeg", "-y", "-i", str(local_file), "-c:v", "libx264", "-preset", "fast", "-crf", "18", "-vf", _vf + ",format=yuv420p", "-colorspace", "bt709", "-color_primaries", "bt709", "-color_trc", "bt709", "-pix_fmt", "yuv420p", "-level", "4.2", ] # 竖屏视频：清除旋转元数据 if _needs_rotation: _cmd.extend(["-metadata:s:v:0", "rotate=0"]) _cmd.extend( [ "-c:a", "aac", "-b:a", "128k", "-movflags", "+faststart", str(_tc_tmp), ] ) _proc = subprocess.run( _cmd, stdout=subprocess.DEVNULL, stderr=subprocess.PIPE, text=True, timeout=900, ) if _proc.returncode == 0 and _tc_tmp.exists() and _tc_tmp.stat().st_size > 0: from video_processing.oss_helpers import upload_to_oss _p = Path(job.storage_key) _new_key = str(_p.parent / (_p.stem + "_h264" + _p.suffix)) _url = upload_to_oss(_tc_tmp, _new_key) if _url: # 先提取元数据，确认成功后再更新 storage_key（避免脏数据） _new_metadata, _new_extract_success = extract_media_metadata( str(_tc_tmp), media_type, ) if _new_extract_success: job.storage_key = _new_key metadata = _new_metadata extract_success = _new_extract_success logger.info( "HEVC→H.264 转码完成: job_id=%s key=%s", job_id, _new_key[:80], ) else: logger.warning( "转码文件上传 OSS 失败，使用原始文件: job_id=%s", job_id, ) else: _tail = _proc.stderr[-300:] if _proc.stderr else "" logger.warning( "FFmpeg 转码失败 rc=%s stderr=%s: job_id=%s", _proc.returncode, _tail, job_id, ) except subprocess.TimeoutExpired: logger.warning( "FFmpeg 转码超时(900s)，降级原始文件: job_id=%s", job_id, ) except Exception as _e: logger.warning( "HEVC 转码异常（降级原始文件）: job_id=%s err=%s", job_id, _e, ) finally: if _tc_tmp and _tc_tmp.exists(): try: _tc_tmp.unlink() except OSError: pass finally: if local_file and local_file.exists(): try: local_file.unlink() except OSError: pass # 有效性校验：ffprobe/Pillow 必须成功，且文件大小/时长/尺寸满足最小要求 is_valid = extract_success and _is_valid_media(metadata, media_type) if not is_valid: # 文件无效，创建 ERROR 状态的 asset 并标记 job 失败 error_reason = "metadata extraction failed" if not extract_success else "media validation failed" logger.warning( "素材有效性校验失败，标记为ERROR: job_id=%s storage_key=%s media_type=%s reason=%s", job_id, job.storage_key, media_type, error_reason, ) asset = Asset.create( project_id=job.project_id, library_id=job.library_id, name=filename, storage_key=job.storage_key, mime_type=mime_type, metadata={"ingest_error": error_reason}, file_size=int(metadata.get("size_bytes", 0)), duration=float(metadata.get("duration", 0)), width=int(metadata.get("width", 0)), height=int(metadata.get("height", 0)), codec=metadata.get("codec") or None, status=AssetStatus.ERROR, file_hash=job.file_hash, ) asset_repo.create(asset) # Update job status to FAILED job.status = IngestJobStatus.FAILED job.error_message = f"Invalid media file: {error_reason}" job.result_asset_id = asset.id job.updated_at = datetime.now(timezone.utc) job_repo.update(job) db.commit() return { "status": "failed", "job_id": job.id, "asset_id": asset.id, "error": error_reason, } # Create Asset asset = Asset.create( project_id=job.project_id, library_id=job.library_id, name=filename, storage_key=job.storage_key, mime_type=mime_type, metadata=metadata, file_size=int(metadata.get("size_bytes", 0)), duration=float(metadata.get("duration", 0)), width=int(metadata.get("width", 0)), height=int(metadata.get("height", 0)), codec=metadata.get("codec") or None, status=AssetStatus.READY, file_hash=job.file_hash, thumbnail_url=thumbnail_url, ) asset_repo.create(asset) # Update job status to COMPLETED job.status = IngestJobStatus.COMPLETED job.result_asset_id = asset.id job.updated_at = datetime.now(timezone.utc) job_repo.update(job) db.commit() return { "status": "completed", "job_id": job.id, "asset_id": asset.id, } except Exception as e: db.rollback() logger.error(f"Failed to ingest asset {job_id}: {e}") # Update job status to FAILED try: job_repo = SQLAlchemyIngestJobRepository(db) job = job_repo.get(job_id) if job: job.status = IngestJobStatus.FAILED job.error_message = str(e) job.updated_at = datetime.now(timezone.utc) job_repo.update(job) db.commit() except Exception: db.rollback() return { "status": "failed", "job_id": job_id, "error": str(e), } finally: db.close()