fix(video): 修复视频定格效果实现逻辑

- 修改cameraShot实现注释，明确标注定格效果功能
- 使用tpad代替freezeframes实现更准确的定格效果
- 更新滤镜链参数配置，确保定格时长正确应用
- 优化变量命名，提高代码可读性
- 调整concat拼接输入源，确保视频流正确连接

.env.example

@@ -1,13 +1,65 @@
-TEMPLATE_DIR=template/
+# ===================
-API_ENDPOINT=https://zhentuai.com/task/v1
+# API 配置
+# ===================
+API_ENDPOINT=http://127.0.0.1:18084/api
 ACCESS_KEY=TEST_ACCESS_KEY
+WORKER_ID=1
+
+# ===================
+# 目录配置
+# ===================
 TEMP_DIR=tmp/
-#REDIRECT_TO_URL=https://renderworker-deuvulkhes.cn-shanghai.fcapp.run/
+
-# QSV
+# ===================
-ENCODER_ARGS="-c:v h264_qsv -global_quality 28 -look_ahead 1"
+# 并发与调度
-# NVENC
+# ===================
-#ENCODER_ARGS="-c:v h264_nvenc -cq:v 24 -preset:v p7 -tune:v hq -profile:v high"
+#MAX_CONCURRENCY=4                    # 最大并发任务数
-# HEVC
+#HEARTBEAT_INTERVAL=5                 # 心跳间隔（秒）
-#VIDEO_ARGS="-profile:v main
+#LEASE_EXTENSION_THRESHOLD=60         # 租约续期阈值（秒），提前多久续期
-UPLOAD_METHOD="rclone"
+#LEASE_EXTENSION_DURATION=300         # 租约续期时长（秒）
-RCLONE_REPLACE_MAP="https://oss.zhentuai.com|alioss://frametour-assets,https://frametour-assets.oss-cn-shanghai.aliyuncs.com|alioss://frametour-assets"
+
+# ===================
+# 能力配置
+# ===================
+# 支持的任务类型，逗号分隔，默认全部支持
+#CAPABILITIES=RENDER_SEGMENT_VIDEO,PREPARE_JOB_AUDIO,PACKAGE_SEGMENT_TS,FINALIZE_MP4
+
+# ===================
+# 超时配置
+# ===================
+#FFMPEG_TIMEOUT=3600                  # FFmpeg 执行超时（秒）
+#DOWNLOAD_TIMEOUT=300                 # 下载超时（秒）
+#UPLOAD_TIMEOUT=600                   # 上传超时（秒）
+
+# ===================
+# 硬件加速与多显卡
+# ===================
+# 硬件加速类型: none, qsv, cuda
+HW_ACCEL=none
+
+# GPU 设备列表（逗号分隔的设备索引）
+# 不配置时：自动检测所有设备
+# 单设备示例：GPU_DEVICES=0
+# 多设备示例：GPU_DEVICES=0,1,2
+#GPU_DEVICES=0,1
+
+# ===================
+# 素材缓存
+# ===================
+#CACHE_ENABLED=true                   # 是否启用素材缓存
+#CACHE_DIR=                           # 缓存目录，默认 TEMP_DIR/cache
+#CACHE_MAX_SIZE_GB=0                  # 最大缓存大小（GB），0 表示不限制
+
+# ===================
+# URL 映射（内网下载加速）
+# ===================
+# 格式: src1|dst1,src2|dst2
+#HTTP_REPLACE_MAP="https://cdcdn.zhentuai.com|http://192.168.10.254:9000"
+
+# ===================
+# 上传配置
+# ===================
+# 上传方式: 默认 HTTP，可选 rclone
+#UPLOAD_METHOD=rclone
+#RCLONE_CONFIG_FILE=                  # rclone 配置文件路径
+#RCLONE_REPLACE_MAP="https://oss.example.com|alioss://bucket"

.gitignore

@@ -31,4 +31,6 @@ target/
 .venv
 venv/
 cython_debug/
-.env
+.env
+.serena
+.claude

app.py

@@ -1,40 +0,0 @@
-import time
-
-import flask
-
-import config
-import biz.task
-import template
-from telemetry import init_opentelemetry
-from template import load_local_template
-from util import api
-
-load_local_template()
-import logging
-
-LOGGER = logging.getLogger(__name__)
-init_opentelemetry(batch=False)
-app = flask.Flask(__name__)
-
-@app.get('/health/check')
-def health_check():
-    return api.sync_center()
-
-@app.post('/')
-def do_nothing():
-    return "NOOP"
-
-@app.post('/<task_id>')
-def do_task(task_id):
-    task_info = api.get_task_info(task_id)
-    local_template_info = template.get_template_def(task_info.get("templateId"))
-    template_info = api.get_template_info(task_info.get("templateId"))
-    if local_template_info:
-        if local_template_info.get("updateTime") != template_info.get("updateTime"):
-            template.download_template(task_info.get("templateId"))
-    biz.task.start_task(task_info)
-    return "OK"
-
-
-if __name__ == '__main__':
-    app.run(host="0.0.0.0", port=9998)

biz/ffmpeg.py

@@ -1,172 +0,0 @@
-import json
-import os.path
-import time
-from concurrent.futures import ThreadPoolExecutor
-
-from opentelemetry.trace import Status, StatusCode
-
-from entity.ffmpeg import FfmpegTask
-import logging
-
-from util import ffmpeg, oss
-from util.ffmpeg import fade_out_audio
-from telemetry import get_tracer
-
-logger = logging.getLogger('biz/ffmpeg')
-
-
-def parse_ffmpeg_task(task_info, template_info):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("parse_ffmpeg_task") as span:
-        tasks = []
-        # 中间片段
-        task_params_str = task_info.get("taskParams", "{}")
-        span.set_attribute("task_params", task_params_str)
-        task_params: dict = json.loads(task_params_str)
-        task_params_orig = json.loads(task_params_str)
-        
-        # 统计only_if占位符的使用次数
-        only_if_usage_count = {}
-        with tracer.start_as_current_span("parse_ffmpeg_task.download_all") as sub_span:
-            with ThreadPoolExecutor(max_workers=8) as executor:
-                param_list: list[dict]
-                for param_list in task_params.values():
-                    for param in param_list:
-                        url = param.get("url")
-                        if url.startswith("http"):
-                            _, fn = os.path.split(url)
-                            executor.submit(oss.download_from_oss, url, fn, True)
-                executor.shutdown(wait=True)
-        for part in template_info.get("video_parts"):
-            source, ext_data = parse_video(part.get('source'), task_params, template_info)
-            if not source:
-                logger.warning("no video found for part: " + str(part))
-                continue
-            only_if = part.get('only_if', '')
-            if only_if:
-                only_if_usage_count[only_if] = only_if_usage_count.get(only_if, 0) + 1
-                required_count = only_if_usage_count.get(only_if)
-                if not check_placeholder_exist_with_count(only_if, task_params_orig, required_count):
-                    logger.info("because only_if exist, placeholder: %s insufficient (need %d), skip part: %s", only_if, required_count, part)
-                    continue
-            sub_ffmpeg_task = FfmpegTask(source)
-            sub_ffmpeg_task.resolution = template_info.get("video_size", "")
-            sub_ffmpeg_task.annexb = True
-            sub_ffmpeg_task.ext_data = ext_data or {}
-            sub_ffmpeg_task.frame_rate = template_info.get("frame_rate", 25)
-            sub_ffmpeg_task.center_cut = part.get("crop_mode", None)
-            sub_ffmpeg_task.zoom_cut = part.get("zoom_cut", None)
-            for effect in part.get('effects', []):
-                sub_ffmpeg_task.add_effect(effect)
-            for lut in part.get('luts', []):
-                sub_ffmpeg_task.add_lut(os.path.join(template_info.get("local_path"), lut).replace("\\", "/"))
-            for audio in part.get('audios', []):
-                sub_ffmpeg_task.add_audios(os.path.join(template_info.get("local_path"), audio))
-            for overlay in part.get('overlays', []):
-                sub_ffmpeg_task.add_overlay(os.path.join(template_info.get("local_path"), overlay))
-            tasks.append(sub_ffmpeg_task)
-        output_file = "out_" + str(time.time()) + ".mp4"
-        task = FfmpegTask(tasks, output_file=output_file)
-        task.resolution = template_info.get("video_size", "")
-        overall = template_info.get("overall_template")
-        task.center_cut = template_info.get("crop_mode", None)
-        task.zoom_cut = template_info.get("zoom_cut", None)
-        task.frame_rate = template_info.get("frame_rate", 25)
-        # if overall.get('source', ''):
-        #     source, ext_data = parse_video(overall.get('source'), task_params, template_info)
-        #     task.add_inputs(source)
-        #     task.ext_data = ext_data or {}
-        for effect in overall.get('effects', []):
-            task.add_effect(effect)
-        for lut in overall.get('luts', []):
-            task.add_lut(os.path.join(template_info.get("local_path"), lut).replace("\\", "/"))
-        for audio in overall.get('audios', []):
-            task.add_audios(os.path.join(template_info.get("local_path"), audio))
-        for overlay in overall.get('overlays', []):
-            task.add_overlay(os.path.join(template_info.get("local_path"), overlay))
-        return task
-
-
-def parse_video(source, task_params, template_info):
-    if source.startswith('PLACEHOLDER_'):
-        placeholder_id = source.replace('PLACEHOLDER_', '')
-        new_sources = task_params.get(placeholder_id, [])
-        _pick_source = {}
-        if type(new_sources) is list:
-            if len(new_sources) == 0:
-                logger.debug("no video found for placeholder: " + placeholder_id)
-                return None, _pick_source
-            else:
-                _pick_source = new_sources.pop(0)
-                new_sources = _pick_source.get("url")
-        if new_sources.startswith("http"):
-            _, source_name = os.path.split(new_sources)
-            oss.download_from_oss(new_sources, source_name, True)
-            return source_name, _pick_source
-        return new_sources, _pick_source
-    return os.path.join(template_info.get("local_path"), source), None
-
-
-def check_placeholder_exist(placeholder_id, task_params):
-    if placeholder_id in task_params:
-        new_sources = task_params.get(placeholder_id, [])
-        if type(new_sources) is list:
-            if len(new_sources) == 0:
-                return False
-            else:
-                return True
-        return True
-    return False
-
-
-def check_placeholder_exist_with_count(placeholder_id, task_params, required_count=1):
-    """检查占位符是否存在足够数量的片段"""
-    if placeholder_id in task_params:
-        new_sources = task_params.get(placeholder_id, [])
-        if type(new_sources) is list:
-            return len(new_sources) >= required_count
-        return required_count <= 1
-    return False
-
-
-def start_ffmpeg_task(ffmpeg_task):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("start_ffmpeg_task") as span:
-        for task in ffmpeg_task.analyze_input_render_tasks():
-            result = start_ffmpeg_task(task)
-            if not result:
-                return False
-        ffmpeg_task.correct_task_type()
-        span.set_attribute("task.type", ffmpeg_task.task_type)
-        span.set_attribute("task.center_cut", str(ffmpeg_task.center_cut))
-        span.set_attribute("task.frame_rate", ffmpeg_task.frame_rate)
-        span.set_attribute("task.resolution", str(ffmpeg_task.resolution))
-        span.set_attribute("task.ext_data", json.dumps(ffmpeg_task.ext_data))
-        result = ffmpeg.start_render(ffmpeg_task)
-        if not result:
-            span.set_status(Status(StatusCode.ERROR))
-            return False
-        span.set_status(Status(StatusCode.OK))
-        return True
-
-
-def clear_task_tmp_file(ffmpeg_task):
-    for task in ffmpeg_task.analyze_input_render_tasks():
-        clear_task_tmp_file(task)
-    try:
-        if os.getenv("TEMPLATE_DIR") not in ffmpeg_task.get_output_file():
-            os.remove(ffmpeg_task.get_output_file())
-            logger.info("delete tmp file: " + ffmpeg_task.get_output_file())
-        else:
-            logger.info("skip delete template file: " + ffmpeg_task.get_output_file())
-    except OSError:
-        logger.warning("delete tmp file failed: " + ffmpeg_task.get_output_file())
-        return False
-    return True
-
-
-def probe_video_info(ffmpeg_task):
-    # 获取视频长度宽度和时长
-    return ffmpeg.probe_video_info(ffmpeg_task.get_output_file())
-
-

biz/task.py

@@ -1,44 +0,0 @@
-import json
-
-from opentelemetry.trace import Status, StatusCode
-
-from biz.ffmpeg import parse_ffmpeg_task, start_ffmpeg_task, clear_task_tmp_file, probe_video_info, fade_out_audio
-from telemetry import get_tracer
-from template import get_template_def
-from util import api
-
-
-def start_task(task_info):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("start_task") as span:
-        task_info = api.normalize_task(task_info)
-        span.set_attribute("task", json.dumps(task_info))
-        span.set_attribute("scenicId", task_info.get("scenicId", "?"))
-        span.set_attribute("templateId", task_info.get("templateId"))
-        template_info = get_template_def(task_info.get("templateId"))
-        api.report_task_start(task_info)
-        ffmpeg_task = parse_ffmpeg_task(task_info, template_info)
-        result = start_ffmpeg_task(ffmpeg_task)
-        if not result:
-            span.set_status(Status(StatusCode.ERROR))
-            return api.report_task_failed(task_info)
-        width, height, duration = probe_video_info(ffmpeg_task)
-        span.set_attribute("probe.width", width)
-        span.set_attribute("probe.height", height)
-        span.set_attribute("probe.duration", duration)
-        # 音频淡出
-        new_fn = fade_out_audio(ffmpeg_task.get_output_file(), duration)
-        ffmpeg_task.set_output_file(new_fn)
-        oss_result = api.upload_task_file(task_info, ffmpeg_task)
-        if not oss_result:
-            span.set_status(Status(StatusCode.ERROR))
-            return api.report_task_failed(task_info)
-        # 获取视频长度宽度和时长
-        clear_task_tmp_file(ffmpeg_task)
-        api.report_task_success(task_info, videoInfo={
-            "width": width,
-            "height": height,
-            "duration": duration
-        })
-        span.set_status(Status(StatusCode.OK))
-        return None

config/__init__.py

@@ -1,16 +0,0 @@
-import datetime
-import logging
-from logging.handlers import TimedRotatingFileHandler
-from dotenv import load_dotenv
-
-load_dotenv()
-logging.basicConfig(level=logging.INFO)
-root_logger = logging.getLogger()
-rf_handler = TimedRotatingFileHandler('all_log.log', when='midnight')
-rf_handler.setFormatter(logging.Formatter("[%(asctime)s][%(name)s]%(levelname)s - %(message)s"))
-rf_handler.setLevel(logging.DEBUG)
-f_handler = TimedRotatingFileHandler('error.log', when='midnight')
-f_handler.setLevel(logging.ERROR)
-f_handler.setFormatter(logging.Formatter("[%(asctime)s][%(name)s][:%(lineno)d]%(levelname)s -  - %(message)s"))
-root_logger.addHandler(rf_handler)
-root_logger.addHandler(f_handler)

constant/__init__.py

@@ -1,9 +1,99 @@
-SUPPORT_FEATURE = (
+# -*- coding: utf-8 -*-
-    'simple_render_algo',
+"""
-    'gpu_accelerate',
+常量定义
-    'hevc_encode',
+
-    'rapid_download',
+v2 版本常量，用于 Render Worker v2 API。
-    'rclone_upload',
+"""
-    'custom_re_encode',
+
+# 软件版本
+SOFTWARE_VERSION = '2.0.0'
+
+# 支持的任务类型
+TASK_TYPES = (
+    'RENDER_SEGMENT_VIDEO',
+    'COMPOSE_TRANSITION',
+    'PREPARE_JOB_AUDIO',
+    'PACKAGE_SEGMENT_TS',
+    'FINALIZE_MP4',
 )
-SOFTWARE_VERSION = '0.0.5'
+
+# 默认能力
+DEFAULT_CAPABILITIES = list(TASK_TYPES)
+
+# 支持的转场类型（对应 FFmpeg xfade 参数）
+TRANSITION_TYPES = (
+    'fade',       # 淡入淡出（默认）
+    'dissolve',   # 溶解过渡
+    'wipeleft',   # 向左擦除
+    'wiperight',  # 向右擦除
+    'wipeup',     # 向上擦除
+    'wipedown',   # 向下擦除
+    'slideleft',  # 向左滑动
+    'slideright', # 向右滑动
+    'slideup',    # 向上滑动
+    'slidedown',  # 向下滑动
+)
+
+# 支持的特效类型
+EFFECT_TYPES = (
+    'cameraShot',  # 相机定格效果：在指定时间点冻结画面
+    'zoom',        # 缩放效果（预留）
+    'blur',        # 模糊效果（预留）
+)
+
+# 硬件加速类型
+HW_ACCEL_NONE = 'none'      # 纯软件编解码
+HW_ACCEL_QSV = 'qsv'        # Intel Quick Sync Video (核显/独显)
+HW_ACCEL_CUDA = 'cuda'      # NVIDIA NVENC/NVDEC
+
+HW_ACCEL_TYPES = (HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA)
+
+# 统一视频编码参数（软件编码，来自集成文档）
+VIDEO_ENCODE_PARAMS = {
+    'codec': 'libx264',
+    'preset': 'medium',
+    'profile': 'main',
+    'level': '4.0',
+    'crf': '23',
+    'pix_fmt': 'yuv420p',
+}
+
+# QSV 硬件加速视频编码参数（Intel Quick Sync）
+VIDEO_ENCODE_PARAMS_QSV = {
+    'codec': 'h264_qsv',
+    'preset': 'medium',          # QSV 支持: veryfast, faster, fast, medium, slow, slower, veryslow
+    'profile': 'main',
+    'level': '4.0',
+    'global_quality': '23',      # QSV 使用 global_quality 代替 crf（1-51，值越低质量越高）
+    'look_ahead': '1',           # 启用前瞻分析提升质量
+    'pix_fmt': 'nv12',           # QSV 硬件表面格式
+}
+
+# CUDA 硬件加速视频编码参数（NVIDIA NVENC）
+VIDEO_ENCODE_PARAMS_CUDA = {
+    'codec': 'h264_nvenc',
+    'preset': 'p4',              # NVENC 预设 p1-p7（p1 最快，p7 最慢/质量最高），p4 ≈ medium
+    'profile': 'main',
+    'level': '4.0',
+    'rc': 'vbr',                 # 码率控制模式：vbr 可变码率
+    'cq': '23',                  # 恒定质量模式的质量值（0-51）
+    'pix_fmt': 'yuv420p',        # NVENC 输入格式（会自动转换）
+}
+
+# 统一音频编码参数
+AUDIO_ENCODE_PARAMS = {
+    'codec': 'aac',
+    'bitrate': '128k',
+    'sample_rate': '48000',
+    'channels': '2',
+}
+
+# 错误码
+ERROR_CODES = {
+    'E_INPUT_UNAVAILABLE': '素材不可访问',
+    'E_FFMPEG_FAILED': 'FFmpeg 执行失败',
+    'E_UPLOAD_FAILED': '上传失败',
+    'E_SPEC_INVALID': '渲染规格非法',
+    'E_TIMEOUT': '执行超时',
+    'E_UNKNOWN': '未知错误',
+}

core/__init__.py

@@ -0,0 +1,12 @@
+# -*- coding: utf-8 -*-
+"""
+核心抽象层
+
+包含任务处理器抽象基类等核心接口定义。
+"""
+
+from core.handler import TaskHandler
+
+__all__ = [
+    'TaskHandler',
+]

core/handler.py

@@ -0,0 +1,79 @@
+# -*- coding: utf-8 -*-
+"""
+任务处理器抽象基类
+
+定义任务处理器的接口规范。
+"""
+
+from abc import ABC, abstractmethod
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from domain.task import Task, TaskType
+    from domain.result import TaskResult
+
+
+class TaskHandler(ABC):
+    """
+    任务处理器抽象基类
+
+    所有任务处理器都必须继承此类并实现相应方法。
+    """
+
+    @abstractmethod
+    def handle(self, task: 'Task') -> 'TaskResult':
+        """
+        处理任务的主方法
+
+        Args:
+            task: 任务实体
+
+        Returns:
+            TaskResult: 任务结果（成功或失败）
+        """
+        pass
+
+    @abstractmethod
+    def get_supported_type(self) -> 'TaskType':
+        """
+        返回此处理器支持的任务类型
+
+        Returns:
+            TaskType: 支持的任务类型枚举值
+        """
+        pass
+
+    def before_handle(self, task: 'Task') -> None:
+        """
+        处理前钩子（可选重写）
+
+        用于任务执行前的准备工作，如日志记录、资源检查等。
+
+        Args:
+            task: 任务实体
+        """
+        pass
+
+    def after_handle(self, task: 'Task', result: 'TaskResult') -> None:
+        """
+        处理后钩子（可选重写）
+
+        用于任务执行后的清理工作，如资源释放、统计记录等。
+
+        Args:
+            task: 任务实体
+            result: 任务结果
+        """
+        pass
+
+    def validate_task(self, task: 'Task') -> bool:
+        """
+        验证任务是否有效（可选重写）
+
+        Args:
+            task: 任务实体
+
+        Returns:
+            bool: 任务是否有效
+        """
+        return True

domain/__init__.py

@@ -0,0 +1,24 @@
+# -*- coding: utf-8 -*-
+"""
+领域模型层
+
+包含任务实体、结果、配置等核心数据结构。
+"""
+
+from domain.task import Task, TaskType, TaskStatus, RenderSpec, OutputSpec, AudioSpec, AudioProfile
+from domain.result import TaskResult, ErrorCode, RETRY_CONFIG
+from domain.config import WorkerConfig
+
+__all__ = [
+    'Task',
+    'TaskType',
+    'TaskStatus',
+    'RenderSpec',
+    'OutputSpec',
+    'AudioSpec',
+    'AudioProfile',
+    'TaskResult',
+    'ErrorCode',
+    'RETRY_CONFIG',
+    'WorkerConfig',
+]

domain/config.py

@@ -0,0 +1,183 @@
+# -*- coding: utf-8 -*-
+"""
+Worker 配置模型
+
+定义 Worker 运行时的配置参数。
+"""
+
+import logging
+import os
+from dataclasses import dataclass, field
+from typing import List, Optional
+
+from constant import HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA, HW_ACCEL_TYPES
+
+logger = logging.getLogger(__name__)
+
+
+# 默认支持的任务类型
+DEFAULT_CAPABILITIES = [
+    "RENDER_SEGMENT_VIDEO",
+    "PREPARE_JOB_AUDIO",
+    "PACKAGE_SEGMENT_TS",
+    "FINALIZE_MP4"
+]
+
+
+@dataclass
+class WorkerConfig:
+    """
+    Worker 配置
+
+    包含 Worker 运行所需的所有配置参数。
+    """
+    # API 配置
+    api_endpoint: str
+    access_key: str
+    worker_id: str
+
+    # 并发控制
+    max_concurrency: int = 4
+
+    # 心跳配置
+    heartbeat_interval: int = 5  # 秒
+
+    # 租约配置
+    lease_extension_threshold: int = 60  # 秒，提前多久续期
+    lease_extension_duration: int = 300  # 秒，每次续期时长
+
+    # 目录配置
+    temp_dir: str = "/tmp/render_worker"
+
+    # 能力配置
+    capabilities: List[str] = field(default_factory=lambda: DEFAULT_CAPABILITIES.copy())
+
+    # FFmpeg 配置
+    ffmpeg_timeout: int = 3600  # 秒，FFmpeg 执行超时
+
+    # 下载/上传配置
+    download_timeout: int = 300  # 秒，下载超时
+    upload_timeout: int = 600  # 秒，上传超时
+
+    # 硬件加速配置
+    hw_accel: str = HW_ACCEL_NONE  # 硬件加速类型: none, qsv, cuda
+
+    # GPU 设备配置（多显卡调度）
+    gpu_devices: List[int] = field(default_factory=list)  # 空列表表示使用默认设备
+
+    # 素材缓存配置
+    cache_enabled: bool = True  # 是否启用素材缓存
+    cache_dir: str = ""  # 缓存目录，默认为 temp_dir/cache
+    cache_max_size_gb: float = 0  # 最大缓存大小（GB），0 表示不限制
+
+    @classmethod
+    def from_env(cls) -> 'WorkerConfig':
+        """从环境变量创建配置"""
+
+        # API 端点，优先使用 V2 版本
+        api_endpoint = os.getenv('API_ENDPOINT_V2') or os.getenv('API_ENDPOINT', '')
+        if not api_endpoint:
+            raise ValueError("API_ENDPOINT_V2 or API_ENDPOINT environment variable is required")
+
+        # Access Key
+        access_key = os.getenv('ACCESS_KEY', '')
+        if not access_key:
+            raise ValueError("ACCESS_KEY environment variable is required")
+
+        # Worker ID
+        worker_id = os.getenv('WORKER_ID', '100001')
+
+        # 并发数
+        max_concurrency = int(os.getenv('MAX_CONCURRENCY', '4'))
+
+        # 心跳间隔
+        heartbeat_interval = int(os.getenv('HEARTBEAT_INTERVAL', '5'))
+
+        # 租约配置
+        lease_extension_threshold = int(os.getenv('LEASE_EXTENSION_THRESHOLD', '60'))
+        lease_extension_duration = int(os.getenv('LEASE_EXTENSION_DURATION', '300'))
+
+        # 临时目录
+        temp_dir = os.getenv('TEMP_DIR', os.getenv('TEMP', '/tmp/render_worker'))
+
+        # 能力列表
+        capabilities_str = os.getenv('CAPABILITIES', '')
+        if capabilities_str:
+            capabilities = [c.strip() for c in capabilities_str.split(',') if c.strip()]
+        else:
+            capabilities = DEFAULT_CAPABILITIES.copy()
+
+        # FFmpeg 超时
+        ffmpeg_timeout = int(os.getenv('FFMPEG_TIMEOUT', '3600'))
+
+        # 下载/上传超时
+        download_timeout = int(os.getenv('DOWNLOAD_TIMEOUT', '300'))
+        upload_timeout = int(os.getenv('UPLOAD_TIMEOUT', '600'))
+
+        # 硬件加速配置
+        hw_accel = os.getenv('HW_ACCEL', HW_ACCEL_NONE).lower()
+        if hw_accel not in HW_ACCEL_TYPES:
+            hw_accel = HW_ACCEL_NONE
+
+        # GPU 设备列表（用于多显卡调度）
+        gpu_devices_str = os.getenv('GPU_DEVICES', '')
+        gpu_devices: List[int] = []
+        if gpu_devices_str:
+            try:
+                gpu_devices = [int(d.strip()) for d in gpu_devices_str.split(',') if d.strip()]
+            except ValueError:
+                logger.warning(f"Invalid GPU_DEVICES value: {gpu_devices_str}, using auto-detect")
+                gpu_devices = []
+
+        # 素材缓存配置
+        cache_enabled = os.getenv('CACHE_ENABLED', 'true').lower() in ('true', '1', 'yes')
+        cache_dir = os.getenv('CACHE_DIR', '')  # 空字符串表示使用默认路径
+        cache_max_size_gb = float(os.getenv('CACHE_MAX_SIZE_GB', '0'))
+
+        return cls(
+            api_endpoint=api_endpoint,
+            access_key=access_key,
+            worker_id=worker_id,
+            max_concurrency=max_concurrency,
+            heartbeat_interval=heartbeat_interval,
+            lease_extension_threshold=lease_extension_threshold,
+            lease_extension_duration=lease_extension_duration,
+            temp_dir=temp_dir,
+            capabilities=capabilities,
+            ffmpeg_timeout=ffmpeg_timeout,
+            download_timeout=download_timeout,
+            upload_timeout=upload_timeout,
+            hw_accel=hw_accel,
+            gpu_devices=gpu_devices,
+            cache_enabled=cache_enabled,
+            cache_dir=cache_dir if cache_dir else os.path.join(temp_dir, 'cache'),
+            cache_max_size_gb=cache_max_size_gb
+        )
+
+    def get_work_dir_path(self, task_id: str) -> str:
+        """获取任务工作目录路径"""
+        return os.path.join(self.temp_dir, f"task_{task_id}")
+
+    def ensure_temp_dir(self) -> None:
+        """确保临时目录存在"""
+        os.makedirs(self.temp_dir, exist_ok=True)
+
+    def is_hw_accel_enabled(self) -> bool:
+        """是否启用了硬件加速"""
+        return self.hw_accel != HW_ACCEL_NONE
+
+    def is_qsv(self) -> bool:
+        """是否使用 QSV 硬件加速"""
+        return self.hw_accel == HW_ACCEL_QSV
+
+    def is_cuda(self) -> bool:
+        """是否使用 CUDA 硬件加速"""
+        return self.hw_accel == HW_ACCEL_CUDA
+
+    def has_multi_gpu(self) -> bool:
+        """是否配置了多 GPU"""
+        return len(self.gpu_devices) > 1
+
+    def get_gpu_devices(self) -> List[int]:
+        """获取 GPU 设备列表"""
+        return self.gpu_devices.copy()

domain/gpu.py

@@ -0,0 +1,31 @@
+# -*- coding: utf-8 -*-
+"""
+GPU 设备模型
+
+定义 GPU 设备的数据结构。
+"""
+
+from dataclasses import dataclass
+from typing import Optional
+
+
+@dataclass
+class GPUDevice:
+    """
+    GPU 设备信息
+
+    Attributes:
+        index: 设备索引（对应 nvidia-smi 中的 GPU ID）
+        name: 设备名称（如 "NVIDIA GeForce RTX 3090"）
+        memory_total: 显存总量（MB），可选
+        available: 设备是否可用
+    """
+    index: int
+    name: str
+    memory_total: Optional[int] = None
+    available: bool = True
+
+    def __str__(self) -> str:
+        status = "available" if self.available else "unavailable"
+        mem_info = f", {self.memory_total}MB" if self.memory_total else ""
+        return f"GPU[{self.index}]: {self.name}{mem_info} ({status})"

domain/result.py

@@ -0,0 +1,105 @@
+# -*- coding: utf-8 -*-
+"""
+任务结果模型
+
+定义错误码、重试配置、任务结果等数据结构。
+"""
+
+from enum import Enum
+from dataclasses import dataclass
+from typing import Optional, Dict, Any, List
+
+
+class ErrorCode(Enum):
+    """错误码枚举"""
+    E_INPUT_UNAVAILABLE = "E_INPUT_UNAVAILABLE"   # 素材不可访问/404
+    E_FFMPEG_FAILED = "E_FFMPEG_FAILED"           # FFmpeg 执行失败
+    E_UPLOAD_FAILED = "E_UPLOAD_FAILED"           # 上传失败
+    E_SPEC_INVALID = "E_SPEC_INVALID"             # renderSpec 非法
+    E_TIMEOUT = "E_TIMEOUT"                        # 执行超时
+    E_UNKNOWN = "E_UNKNOWN"                        # 未知错误
+
+
+# 重试配置
+RETRY_CONFIG: Dict[ErrorCode, Dict[str, Any]] = {
+    ErrorCode.E_INPUT_UNAVAILABLE: {
+        'max_retries': 3,
+        'backoff': [1, 2, 5]  # 重试间隔（秒）
+    },
+    ErrorCode.E_FFMPEG_FAILED: {
+        'max_retries': 2,
+        'backoff': [1, 3]
+    },
+    ErrorCode.E_UPLOAD_FAILED: {
+        'max_retries': 3,
+        'backoff': [1, 2, 5]
+    },
+    ErrorCode.E_SPEC_INVALID: {
+        'max_retries': 0,  # 不重试
+        'backoff': []
+    },
+    ErrorCode.E_TIMEOUT: {
+        'max_retries': 2,
+        'backoff': [5, 10]
+    },
+    ErrorCode.E_UNKNOWN: {
+        'max_retries': 1,
+        'backoff': [2]
+    },
+}
+
+
+@dataclass
+class TaskResult:
+    """
+    任务结果
+
+    封装任务执行的结果，包括成功数据或失败信息。
+    """
+    success: bool
+    data: Optional[Dict[str, Any]] = None
+    error_code: Optional[ErrorCode] = None
+    error_message: Optional[str] = None
+
+    @classmethod
+    def ok(cls, data: Dict[str, Any]) -> 'TaskResult':
+        """创建成功结果"""
+        return cls(success=True, data=data)
+
+    @classmethod
+    def fail(cls, error_code: ErrorCode, error_message: str) -> 'TaskResult':
+        """创建失败结果"""
+        return cls(
+            success=False,
+            error_code=error_code,
+            error_message=error_message
+        )
+
+    def to_report_dict(self) -> Dict[str, Any]:
+        """
+        转换为上报格式
+
+        用于 API 上报时的数据格式转换。
+        """
+        if self.success:
+            return {'result': self.data}
+        else:
+            return {
+                'errorCode': self.error_code.value if self.error_code else 'E_UNKNOWN',
+                'errorMessage': self.error_message or 'Unknown error'
+            }
+
+    def can_retry(self) -> bool:
+        """是否可以重试"""
+        if self.success:
+            return False
+        if not self.error_code:
+            return True
+        config = RETRY_CONFIG.get(self.error_code, {})
+        return config.get('max_retries', 0) > 0
+
+    def get_retry_config(self) -> Dict[str, Any]:
+        """获取重试配置"""
+        if not self.error_code:
+            return {'max_retries': 1, 'backoff': [2]}
+        return RETRY_CONFIG.get(self.error_code, {'max_retries': 1, 'backoff': [2]})

domain/task.py

@@ -0,0 +1,554 @@
+# -*- coding: utf-8 -*-
+"""
+任务领域模型
+
+定义任务类型、任务实体、渲染规格、输出规格等数据结构。
+"""
+
+from enum import Enum
+from dataclasses import dataclass, field
+from typing import Dict, Any, Optional, List
+from datetime import datetime
+from urllib.parse import urlparse, unquote
+import os
+
+
+# 支持的图片扩展名
+IMAGE_EXTENSIONS = {'.jpg', '.jpeg', '.png', '.webp', '.bmp', '.gif'}
+
+
+class TaskType(Enum):
+    """任务类型枚举"""
+    RENDER_SEGMENT_VIDEO = "RENDER_SEGMENT_VIDEO"  # 渲染视频片段
+    COMPOSE_TRANSITION = "COMPOSE_TRANSITION"      # 合成转场效果
+    PREPARE_JOB_AUDIO = "PREPARE_JOB_AUDIO"        # 生成全局音频
+    PACKAGE_SEGMENT_TS = "PACKAGE_SEGMENT_TS"      # 封装 TS 分片
+    FINALIZE_MP4 = "FINALIZE_MP4"                  # 产出最终 MP4
+
+
+# 支持的转场类型（对应 FFmpeg xfade 参数）
+TRANSITION_TYPES = {
+    'fade': 'fade',           # 淡入淡出（默认）
+    'dissolve': 'dissolve',   # 溶解过渡
+    'wipeleft': 'wipeleft',   # 向左擦除
+    'wiperight': 'wiperight', # 向右擦除
+    'wipeup': 'wipeup',       # 向上擦除
+    'wipedown': 'wipedown',   # 向下擦除
+    'slideleft': 'slideleft', # 向左滑动
+    'slideright': 'slideright', # 向右滑动
+    'slideup': 'slideup',     # 向上滑动
+    'slidedown': 'slidedown', # 向下滑动
+}
+
+# 支持的特效类型
+EFFECT_TYPES = {
+    'cameraShot',  # 相机定格效果
+    'zoom',        # 缩放效果（预留）
+    'blur',        # 模糊效果（预留）
+}
+
+
+class TaskStatus(Enum):
+    """任务状态枚举"""
+    PENDING = "PENDING"
+    RUNNING = "RUNNING"
+    SUCCESS = "SUCCESS"
+    FAILED = "FAILED"
+
+
+@dataclass
+class TransitionConfig:
+    """
+    转场配置
+
+    用于 RENDER_SEGMENT_VIDEO 任务的入场/出场转场配置。
+    """
+    type: str = "fade"       # 转场类型
+    duration_ms: int = 500   # 转场时长（毫秒）
+
+    @classmethod
+    def from_dict(cls, data: Optional[Dict]) -> Optional['TransitionConfig']:
+        """从字典创建 TransitionConfig"""
+        if not data:
+            return None
+        trans_type = data.get('type', 'fade')
+        # 验证转场类型是否支持
+        if trans_type not in TRANSITION_TYPES:
+            trans_type = 'fade'
+        return cls(
+            type=trans_type,
+            duration_ms=int(data.get('durationMs', 500))
+        )
+
+    def get_overlap_ms(self) -> int:
+        """获取 overlap 时长（单边，为转场时长的一半）"""
+        return self.duration_ms // 2
+
+    def get_ffmpeg_transition(self) -> str:
+        """获取 FFmpeg xfade 参数"""
+        return TRANSITION_TYPES.get(self.type, 'fade')
+
+
+@dataclass
+class Effect:
+    """
+    特效配置
+
+    格式：type:params
+    例如：cameraShot:3,1 表示在第3秒定格1秒
+    """
+    effect_type: str           # 效果类型
+    params: str = ""           # 参数字符串
+
+    @classmethod
+    def from_string(cls, effect_str: str) -> Optional['Effect']:
+        """
+        从字符串解析 Effect
+
+        格式：type:params 或 type（无参数时）
+        """
+        if not effect_str:
+            return None
+        parts = effect_str.split(':', 1)
+        effect_type = parts[0].strip()
+        if effect_type not in EFFECT_TYPES:
+            return None
+        params = parts[1].strip() if len(parts) > 1 else ""
+        return cls(effect_type=effect_type, params=params)
+
+    @classmethod
+    def parse_effects(cls, effects_str: Optional[str]) -> List['Effect']:
+        """
+        解析效果字符串
+
+        格式：effect1|effect2|effect3
+        例如：cameraShot:3,1|blur:5
+        """
+        if not effects_str:
+            return []
+        effects = []
+        for part in effects_str.split('|'):
+            effect = cls.from_string(part.strip())
+            if effect:
+                effects.append(effect)
+        return effects
+
+    def get_camera_shot_params(self) -> tuple:
+        """
+        获取 cameraShot 效果参数
+
+        Returns:
+            (start_sec, duration_sec): 开始时间和持续时间（秒）
+        """
+        if self.effect_type != 'cameraShot':
+            return (0, 0)
+        if not self.params:
+            return (3, 1)  # 默认值
+        parts = self.params.split(',')
+        try:
+            start = int(parts[0]) if len(parts) >= 1 else 3
+            duration = int(parts[1]) if len(parts) >= 2 else 1
+            return (start, duration)
+        except ValueError:
+            return (3, 1)
+
+
+@dataclass
+class RenderSpec:
+    """
+    渲染规格
+
+    用于 RENDER_SEGMENT_VIDEO 任务，定义视频渲染参数。
+    """
+    crop_enable: bool = False
+    crop_size: Optional[str] = None
+    speed: str = "1.0"
+    lut_url: Optional[str] = None
+    overlay_url: Optional[str] = None
+    effects: Optional[str] = None
+    zoom_cut: bool = False
+    video_crop: Optional[str] = None
+    face_pos: Optional[str] = None
+    transitions: Optional[str] = None
+    # 转场配置（PRD v2 新增）
+    transition_in: Optional[TransitionConfig] = None   # 入场转场
+    transition_out: Optional[TransitionConfig] = None  # 出场转场
+
+    @classmethod
+    def from_dict(cls, data: Optional[Dict]) -> 'RenderSpec':
+        """从字典创建 RenderSpec"""
+        if not data:
+            return cls()
+        return cls(
+            crop_enable=data.get('cropEnable', False),
+            crop_size=data.get('cropSize'),
+            speed=str(data.get('speed', '1.0')),
+            lut_url=data.get('lutUrl'),
+            overlay_url=data.get('overlayUrl'),
+            effects=data.get('effects'),
+            zoom_cut=data.get('zoomCut', False),
+            video_crop=data.get('videoCrop'),
+            face_pos=data.get('facePos'),
+            transitions=data.get('transitions'),
+            transition_in=TransitionConfig.from_dict(data.get('transitionIn')),
+            transition_out=TransitionConfig.from_dict(data.get('transitionOut'))
+        )
+
+    def has_transition_in(self) -> bool:
+        """是否有入场转场"""
+        return self.transition_in is not None and self.transition_in.duration_ms > 0
+
+    def has_transition_out(self) -> bool:
+        """是否有出场转场"""
+        return self.transition_out is not None and self.transition_out.duration_ms > 0
+
+    def get_overlap_head_ms(self) -> int:
+        """获取头部 overlap 时长（毫秒）"""
+        if self.has_transition_in():
+            return self.transition_in.get_overlap_ms()
+        return 0
+
+    def get_overlap_tail_ms(self) -> int:
+        """获取尾部 overlap 时长（毫秒）"""
+        if self.has_transition_out():
+            return self.transition_out.get_overlap_ms()
+        return 0
+
+    def get_effects(self) -> List['Effect']:
+        """获取解析后的特效列表"""
+        return Effect.parse_effects(self.effects)
+
+
+@dataclass
+class OutputSpec:
+    """
+    输出规格
+
+    用于 RENDER_SEGMENT_VIDEO 任务，定义视频输出参数。
+    """
+    width: int = 1080
+    height: int = 1920
+    fps: int = 30
+    bitrate: int = 4000000
+    codec: str = "h264"
+
+    @classmethod
+    def from_dict(cls, data: Optional[Dict]) -> 'OutputSpec':
+        """从字典创建 OutputSpec"""
+        if not data:
+            return cls()
+        return cls(
+            width=data.get('width', 1080),
+            height=data.get('height', 1920),
+            fps=data.get('fps', 30),
+            bitrate=data.get('bitrate', 4000000),
+            codec=data.get('codec', 'h264')
+        )
+
+
+@dataclass
+class AudioSpec:
+    """
+    音频规格
+
+    用于 PREPARE_JOB_AUDIO 任务中的片段叠加音效。
+    """
+    audio_url: Optional[str] = None
+    volume: float = 1.0
+    fade_in_ms: int = 10
+    fade_out_ms: int = 10
+    start_ms: int = 0
+    delay_ms: int = 0
+    loop_enable: bool = False
+
+    @classmethod
+    def from_dict(cls, data: Optional[Dict]) -> Optional['AudioSpec']:
+        """从字典创建 AudioSpec"""
+        if not data:
+            return None
+        return cls(
+            audio_url=data.get('audioUrl'),
+            volume=float(data.get('volume', 1.0)),
+            fade_in_ms=int(data.get('fadeInMs', 10)),
+            fade_out_ms=int(data.get('fadeOutMs', 10)),
+            start_ms=int(data.get('startMs', 0)),
+            delay_ms=int(data.get('delayMs', 0)),
+            loop_enable=data.get('loopEnable', False)
+        )
+
+
+@dataclass
+class AudioProfile:
+    """
+    音频配置
+
+    用于 PREPARE_JOB_AUDIO 任务的全局音频参数。
+    """
+    sample_rate: int = 48000
+    channels: int = 2
+    codec: str = "aac"
+
+    @classmethod
+    def from_dict(cls, data: Optional[Dict]) -> 'AudioProfile':
+        """从字典创建 AudioProfile"""
+        if not data:
+            return cls()
+        return cls(
+            sample_rate=data.get('sampleRate', 48000),
+            channels=data.get('channels', 2),
+            codec=data.get('codec', 'aac')
+        )
+
+
+@dataclass
+class Task:
+    """
+    任务实体
+
+    表示一个待执行的渲染任务。
+    """
+    task_id: str
+    task_type: TaskType
+    priority: int
+    lease_expire_time: datetime
+    payload: Dict[str, Any]
+
+    @classmethod
+    def from_dict(cls, data: Dict) -> 'Task':
+        """从 API 响应字典创建 Task"""
+        lease_time_str = data.get('leaseExpireTime', '')
+
+        # 解析 ISO 8601 时间格式
+        if lease_time_str:
+            if lease_time_str.endswith('Z'):
+                lease_time_str = lease_time_str[:-1] + '+00:00'
+            try:
+                lease_expire_time = datetime.fromisoformat(lease_time_str)
+            except ValueError:
+                # 解析失败时使用当前时间 + 5分钟
+                lease_expire_time = datetime.now()
+        else:
+            lease_expire_time = datetime.now()
+
+        return cls(
+            task_id=str(data['taskId']),
+            task_type=TaskType(data['taskType']),
+            priority=data.get('priority', 0),
+            lease_expire_time=lease_expire_time,
+            payload=data.get('payload', {})
+        )
+
+    def get_job_id(self) -> str:
+        """获取作业 ID"""
+        return str(self.payload.get('jobId', ''))
+
+    def get_segment_id(self) -> Optional[str]:
+        """获取片段 ID（如果有）"""
+        segment_id = self.payload.get('segmentId')
+        return str(segment_id) if segment_id else None
+
+    def get_plan_segment_index(self) -> int:
+        """获取计划片段索引"""
+        return int(self.payload.get('planSegmentIndex', 0))
+
+    def get_duration_ms(self) -> int:
+        """获取时长（毫秒）"""
+        return int(self.payload.get('durationMs', 5000))
+
+    def get_material_url(self) -> Optional[str]:
+        """
+        获取素材 URL
+
+        优先使用 boundMaterialUrl（实际可下载的 HTTP URL），
+        如果不存在则回退到 sourceRef（可能是 slot 引用）。
+
+        Returns:
+            素材 URL，如果都不存在返回 None
+        """
+        return self.payload.get('boundMaterialUrl') or self.payload.get('sourceRef')
+
+    def get_source_ref(self) -> Optional[str]:
+        """获取素材源引用（slot 标识符，如 device:xxx）"""
+        return self.payload.get('sourceRef')
+
+    def get_bound_material_url(self) -> Optional[str]:
+        """获取绑定的素材 URL（实际可下载的 HTTP URL）"""
+        return self.payload.get('boundMaterialUrl')
+
+    def get_material_type(self) -> str:
+        """
+        获取素材类型
+
+        优先使用服务端下发的 materialType 字段，
+        如果不存在则根据 URL 后缀自动推断。
+
+        Returns:
+            素材类型："video" 或 "image"
+        """
+        # 优先使用服务端下发的类型
+        material_type = self.payload.get('materialType')
+        if material_type in ('video', 'image'):
+            return material_type
+
+        # 降级：根据 URL 后缀推断
+        material_url = self.get_material_url()
+        if material_url:
+            parsed = urlparse(material_url)
+            path = unquote(parsed.path)
+            _, ext = os.path.splitext(path)
+            if ext.lower() in IMAGE_EXTENSIONS:
+                return 'image'
+
+        # 默认视频类型
+        return 'video'
+
+    def is_image_material(self) -> bool:
+        """判断素材是否为图片类型"""
+        return self.get_material_type() == 'image'
+
+    def get_render_spec(self) -> RenderSpec:
+        """获取渲染规格"""
+        return RenderSpec.from_dict(self.payload.get('renderSpec'))
+
+    def get_output_spec(self) -> OutputSpec:
+        """获取输出规格"""
+        return OutputSpec.from_dict(self.payload.get('output'))
+
+    def get_transition_type(self) -> Optional[str]:
+        """获取转场类型（来自 TaskPayload 顶层）"""
+        return self.payload.get('transitionType')
+
+    def get_transition_ms(self) -> int:
+        """获取转场时长（毫秒，来自 TaskPayload 顶层）"""
+        return int(self.payload.get('transitionMs', 0))
+
+    def has_transition(self) -> bool:
+        """是否有转场效果"""
+        return self.get_transition_ms() > 0
+
+    def get_overlap_tail_ms(self) -> int:
+        """
+        获取尾部 overlap 时长（毫秒）
+
+        转场发生在当前片段与下一片段之间，当前片段需要在尾部多渲染 overlap 帧。
+        overlap = transitionMs / 2
+        """
+        return self.get_transition_ms() // 2
+
+    def get_transition_in_type(self) -> Optional[str]:
+        """获取入场转场类型（来自前一片段的出场转场）"""
+        return self.payload.get('transitionInType')
+
+    def get_transition_in_ms(self) -> int:
+        """获取入场转场时长（毫秒）"""
+        return int(self.payload.get('transitionInMs', 0))
+
+    def get_transition_out_type(self) -> Optional[str]:
+        """获取出场转场类型（当前片段的转场配置）"""
+        return self.payload.get('transitionOutType')
+
+    def get_transition_out_ms(self) -> int:
+        """获取出场转场时长（毫秒）"""
+        return int(self.payload.get('transitionOutMs', 0))
+
+    def has_transition_in(self) -> bool:
+        """是否有入场转场"""
+        return self.get_transition_in_ms() > 0
+
+    def has_transition_out(self) -> bool:
+        """是否有出场转场"""
+        return self.get_transition_out_ms() > 0
+
+    def get_overlap_head_ms(self) -> int:
+        """
+        获取头部 overlap 时长（毫秒）
+
+        入场转场来自前一个片段，当前片段需要在头部多渲染 overlap 帧。
+        overlap = transitionInMs / 2
+        """
+        return self.get_transition_in_ms() // 2
+
+    def get_overlap_tail_ms_v2(self) -> int:
+        """
+        获取尾部 overlap 时长（毫秒）- 使用新的字段名
+
+        出场转场用于当前片段与下一片段之间，当前片段需要在尾部多渲染 overlap 帧。
+        overlap = transitionOutMs / 2
+        """
+        return self.get_transition_out_ms() // 2
+
+    def get_bgm_url(self) -> Optional[str]:
+        """获取 BGM URL"""
+        return self.payload.get('bgmUrl')
+
+    def get_total_duration_ms(self) -> int:
+        """获取总时长（毫秒）"""
+        return int(self.payload.get('totalDurationMs', 0))
+
+    def get_segments(self) -> List[Dict]:
+        """获取片段列表"""
+        return self.payload.get('segments', [])
+
+    def get_audio_profile(self) -> AudioProfile:
+        """获取音频配置"""
+        return AudioProfile.from_dict(self.payload.get('audioProfile'))
+
+    def get_video_url(self) -> Optional[str]:
+        """获取视频 URL（用于 PACKAGE_SEGMENT_TS）"""
+        return self.payload.get('videoUrl')
+
+    def get_audio_url(self) -> Optional[str]:
+        """获取音频 URL（用于 PACKAGE_SEGMENT_TS）"""
+        return self.payload.get('audioUrl')
+
+    def get_start_time_ms(self) -> int:
+        """获取开始时间（毫秒）"""
+        return int(self.payload.get('startTimeMs', 0))
+
+    def get_m3u8_url(self) -> Optional[str]:
+        """获取 m3u8 URL（用于 FINALIZE_MP4）"""
+        return self.payload.get('m3u8Url')
+
+    def get_ts_list(self) -> List[str]:
+        """获取 TS 列表（用于 FINALIZE_MP4）"""
+        return self.payload.get('tsList', [])
+
+    # ========== COMPOSE_TRANSITION 相关方法 ==========
+
+    def get_transition_id(self) -> Optional[str]:
+        """获取转场 ID（用于 COMPOSE_TRANSITION）"""
+        return self.payload.get('transitionId')
+
+    def get_prev_segment(self) -> Optional[Dict]:
+        """获取前一个片段信息（用于 COMPOSE_TRANSITION）"""
+        return self.payload.get('prevSegment')
+
+    def get_next_segment(self) -> Optional[Dict]:
+        """获取后一个片段信息（用于 COMPOSE_TRANSITION）"""
+        return self.payload.get('nextSegment')
+
+    def get_transition_config(self) -> Optional[TransitionConfig]:
+        """获取转场配置（用于 COMPOSE_TRANSITION）"""
+        return TransitionConfig.from_dict(self.payload.get('transition'))
+
+    # ========== PACKAGE_SEGMENT_TS 转场相关方法 ==========
+
+    def is_transition_segment(self) -> bool:
+        """是否为转场分片（用于 PACKAGE_SEGMENT_TS）"""
+        return self.payload.get('isTransitionSegment', False)
+
+    def should_trim_head(self) -> bool:
+        """是否需要裁剪头部 overlap（用于 PACKAGE_SEGMENT_TS）"""
+        return self.payload.get('trimHead', False)
+
+    def should_trim_tail(self) -> bool:
+        """是否需要裁剪尾部 overlap（用于 PACKAGE_SEGMENT_TS）"""
+        return self.payload.get('trimTail', False)
+
+    def get_trim_head_ms(self) -> int:
+        """获取头部裁剪时长（毫秒）"""
+        return int(self.payload.get('trimHeadMs', 0))
+
+    def get_trim_tail_ms(self) -> int:
+        """获取尾部裁剪时长（毫秒）"""
+        return int(self.payload.get('trimTailMs', 0))

entity/ffmpeg.py

@@ -1,507 +0,0 @@
-import json
-import os
-import time
-import uuid
-from typing import Any
-
-DEFAULT_ARGS = ("-shortest",)
-ENCODER_ARGS = ("-c:v", "h264", ) if not os.getenv("ENCODER_ARGS", False) else os.getenv("ENCODER_ARGS", "").split(" ")
-VIDEO_ARGS = ("-profile:v", "high", "-level:v", "4", ) if not os.getenv("VIDEO_ARGS", False) else os.getenv("VIDEO_ARGS", "").split(" ")
-AUDIO_ARGS = ("-c:a", "aac", "-b:a", "128k", "-ar", "48000", "-ac", "2", )
-MUTE_AUDIO_INPUT = ("-f", "lavfi", "-i", "anullsrc=cl=stereo:r=48000", )
-
-
-def get_mp4toannexb_filter():
-    """
-    Determine which mp4toannexb filter to use based on ENCODER_ARGS.
-    Returns 'hevc_mp4toannexb' if ENCODER_ARGS contains 'hevc', otherwise 'h264_mp4toannexb'.
-    """
-    encoder_args_str = os.getenv("ENCODER_ARGS", "").lower()
-    if "hevc" in encoder_args_str:
-        return "hevc_mp4toannexb"
-    return "h264_mp4toannexb"
-
-
-class FfmpegTask(object):
-
-    effects: list[str]
-
-    def __init__(self, input_file, task_type='copy', output_file=''):
-        self.annexb = False
-        if type(input_file) is str:
-            if input_file.endswith(".ts"):
-                self.annexb = True
-            self.input_file = [input_file]
-        elif type(input_file) is list:
-            self.input_file = input_file
-        else:
-            self.input_file = []
-        self.zoom_cut = None
-        self.center_cut = None
-        self.ext_data = {}
-        self.task_type = task_type
-        self.output_file = output_file
-        self.mute = True
-        self.speed = 1
-        self.frame_rate = 25
-        self.resolution = None
-        self.subtitles = []
-        self.luts = []
-        self.audios = []
-        self.overlays = []
-        self.effects = []
-
-    def __repr__(self):
-        _str = f'FfmpegTask(input_file={self.input_file}, task_type={self.task_type}'
-        if len(self.luts) > 0:
-            _str += f', luts={self.luts}'
-        if len(self.audios) > 0:
-            _str += f', audios={self.audios}'
-        if len(self.overlays) > 0:
-            _str += f', overlays={self.overlays}'
-        if self.annexb:
-            _str += f', annexb={self.annexb}'
-        if self.effects:
-            _str += f', effects={self.effects}'
-        if self.mute:
-            _str += f', mute={self.mute}'
-        _str += f', center_cut={self.center_cut}'
-        return _str + ')'
-
-    def analyze_input_render_tasks(self):
-        for i in self.input_file:
-            if type(i) is str:
-                continue
-            elif isinstance(i, FfmpegTask):
-                if i.need_run():
-                    yield i
-
-    def need_run(self):
-        """
-        判断是否需要运行
-        :rtype: bool
-        :return:
-        """
-        if self.annexb:
-            return True
-        # TODO: copy from url
-        return not self.check_can_copy()
-
-    def add_inputs(self, *inputs):
-        self.input_file.extend(inputs)
-
-    def add_overlay(self, *overlays):
-        for overlay in overlays:
-            if str(overlay).endswith('.ass'):
-                self.subtitles.append(overlay)
-            else:
-                self.overlays.append(overlay)
-        self.correct_task_type()
-
-    def add_audios(self, *audios):
-        self.audios.extend(audios)
-        self.correct_task_type()
-        self.check_audio_track()
-
-    def add_lut(self, *luts):
-        self.luts.extend(luts)
-        self.correct_task_type()
-
-    def add_effect(self, *effects):
-        self.effects.extend(effects)
-        self.correct_task_type()
-
-    def get_output_file(self):
-        if self.task_type == 'copy':
-            return self.input_file[0]
-        if self.output_file == '':
-            self.set_output_file()
-        return self.output_file
-
-    def correct_task_type(self):
-        if self.check_can_copy():
-            self.task_type = 'copy'
-        elif self.check_can_concat():
-            self.task_type = 'concat'
-        else:
-            self.task_type = 'encode'
-
-    def check_can_concat(self):
-        if len(self.luts) > 0:
-            return False
-        if len(self.overlays) > 0:
-            return False
-        if len(self.subtitles) > 0:
-            return False
-        if len(self.effects) > 0:
-            return False
-        if self.speed != 1:
-            return False
-        if self.zoom_cut is not None:
-            return False
-        if self.center_cut is not None:
-            return False
-        return True
-
-    def check_can_copy(self):
-        if len(self.luts) > 0:
-            return False
-        if len(self.overlays) > 0:
-            return False
-        if len(self.subtitles) > 0:
-            return False
-        if len(self.effects) > 0:
-            return False
-        if self.speed != 1:
-            return False
-        if len(self.audios) >= 1:
-            return False
-        if len(self.input_file) > 1:
-            return False
-        if self.zoom_cut is not None:
-            return False
-        if self.center_cut is not None:
-            return False
-        return True
-
-    def check_audio_track(self):
-        ...
-
-    def get_ffmpeg_args(self):
-        args = ['-y', '-hide_banner']
-        if self.task_type == 'encode':
-            input_args = []
-            filter_args = []
-            output_args = [*VIDEO_ARGS, *AUDIO_ARGS, *ENCODER_ARGS, *DEFAULT_ARGS]
-            if self.annexb:
-                output_args.append("-bsf:v")
-                output_args.append(get_mp4toannexb_filter())
-                output_args.append("-reset_timestamps")
-                output_args.append("1")
-            video_output_str = "[0:v]"
-            audio_output_str = ""
-            audio_track_index = 0
-            effect_index = 0
-            for input_file in self.input_file:
-                input_args.append("-i")
-                if type(input_file) is str:
-                    input_args.append(input_file)
-                elif isinstance(input_file, FfmpegTask):
-                    input_args.append(input_file.get_output_file())
-            if self.center_cut == 1:
-                pos_json_str = self.ext_data.get('posJson', '{}')
-                try:
-                    pos_json = json.loads(pos_json_str)
-                except Exception as e:
-                    pos_json = {}
-                _v_w = pos_json.get('imgWidth', 1)
-                _f_x = pos_json.get('ltX', 0)
-                _f_x2 = pos_json.get('rbX', 0)
-                _x = f'{float((_f_x2 + _f_x)/(2 * _v_w)) :.4f}*iw-ih*ih/(2*iw)'
-                filter_args.append(f"{video_output_str}crop=x={_x}:y=0:w=ih*ih/iw:h=ih[v_cut{effect_index}]")
-                video_output_str = f"[v_cut{effect_index}]"
-                effect_index += 1
-            if self.zoom_cut == 1 and self.resolution:
-                _input = None
-                for input_file in self.input_file:
-                    if type(input_file) is str:
-                        _input = input_file
-                        break
-                    elif isinstance(input_file, FfmpegTask):
-                        _input = input_file.get_output_file()
-                        break
-                if _input:
-                    from util.ffmpeg import probe_video_info
-                    _iw, _ih, _ = probe_video_info(_input)
-                    _w, _h = self.resolution.split('x', 1)
-                    pos_json_str = self.ext_data.get('posJson', '{}')
-                    try:
-                        pos_json = json.loads(pos_json_str)
-                    except Exception as e:
-                        pos_json = {}
-                    _v_w = pos_json.get('imgWidth', 1)
-                    _v_h = pos_json.get('imgHeight', 1)
-                    _f_x = pos_json.get('ltX', 0)
-                    _f_x2 = pos_json.get('rbX', 0)
-                    _f_y = pos_json.get('ltY', 0)
-                    _f_y2 = pos_json.get('rbY', 0)
-                    _x = min(max(0, int((_f_x + _f_x2) / 2 - int(_w) / 2)), _iw - int(_w))
-                    _y = min(max(0, int((_f_y + _f_y2) / 2 - int(_h) / 2)), _ih - int(_h))
-                    filter_args.append(f"{video_output_str}crop=x={_x}:y={_y}:w={_w}:h={_h}[vz_cut{effect_index}]")
-                    video_output_str = f"[vz_cut{effect_index}]"
-                    effect_index += 1
-            for effect in self.effects:
-                if effect.startswith("cameraShot:"):
-                    param = effect.split(":", 2)[1]
-                    if param == '':
-                        param = "3,1,0"
-                    _split = param.split(",")
-                    start = 3
-                    duration = 1
-                    rotate_deg = 0
-                    if len(_split) >= 3:
-                        if _split[2] == '':
-                            rotate_deg = 0
-                        else:
-                            rotate_deg = int(_split[2])
-                    if len(_split) >= 2:
-                        duration = float(_split[1])
-                    if len(_split) >= 1:
-                        start = float(_split[0])
-                    _start_out_str = "[eff_s]"
-                    _mid_out_str = "[eff_m]"
-                    _end_out_str = "[eff_e]"
-                    filter_args.append(f"{video_output_str}split=3{_start_out_str}{_mid_out_str}{_end_out_str}")
-                    filter_args.append(f"{_start_out_str}select=lt(n\\,{int(start * self.frame_rate)}){_start_out_str}")
-                    filter_args.append(f"{_end_out_str}select=gt(n\\,{int(start * self.frame_rate)}){_end_out_str}")
-                    filter_args.append(f"{_mid_out_str}select=eq(n\\,{int(start * self.frame_rate)}){_mid_out_str}")
-                    filter_args.append(
-                        f"{_mid_out_str}tpad=start_mode=clone:start_duration={duration:.4f}{_mid_out_str}")
-                    if rotate_deg != 0:
-                        filter_args.append(f"{_mid_out_str}rotate=PI*{rotate_deg}/180{_mid_out_str}")
-                    # filter_args.append(f"{video_output_str}trim=start=0:end={start+duration},tpad=stop_mode=clone:stop_duration={duration},setpts=PTS-STARTPTS{_start_out_str}")
-                    # filter_args.append(f"tpad=start_mode=clone:start_duration={duration},setpts=PTS-STARTPTS{_start_out_str}")
-                    # filter_args.append(f"{_end_out_str}trim=start={start}{_end_out_str}")
-                    video_output_str = f"[v_eff{effect_index}]"
-                    # filter_args.append(f"{_end_out_str}{_start_out_str}overlay=eof_action=pass{video_output_str}")
-                    filter_args.append(f"{_start_out_str}{_mid_out_str}{_end_out_str}concat=n=3:v=1:a=0,setpts=N/{self.frame_rate}/TB{video_output_str}")
-                    effect_index += 1
-                elif effect.startswith("ospeed:"):
-                    param = effect.split(":", 2)[1]
-                    if param == '':
-                        param = "1"
-                    if param != "1":
-                        # 视频变速
-                        effect_index += 1
-                        filter_args.append(f"{video_output_str}setpts={param}*PTS[v_eff{effect_index}]")
-                        video_output_str = f"[v_eff{effect_index}]"
-                elif effect.startswith("zoom:"):
-                    param = effect.split(":", 2)[1]
-                    if param == '':
-                        continue
-                    _split = param.split(",")
-                    if len(_split) < 3:
-                        continue
-                    try:
-                        start_time = float(_split[0])
-                        zoom_factor = float(_split[1])
-                        duration = float(_split[2])
-                        if start_time < 0:
-                            start_time = 0
-                        if duration < 0:
-                            duration = 0
-                        if zoom_factor <= 0:
-                            zoom_factor = 1
-                    except (ValueError, IndexError):
-                        start_time = 0
-                        duration = 0
-                        zoom_factor = 1
-                    if zoom_factor == 1:
-                        continue
-                    effect_index += 1
-                    
-                    # 获取缩放中心点（从pos_json或使用默认中心）
-                    center_x = "iw/2"
-                    center_y = "ih/2"
-                    pos_json_str = self.ext_data.get('posJson', '{}')
-                    try:
-                        pos_json = json.loads(pos_json_str) if pos_json_str != '{}' else {}
-                        if pos_json:
-                            _f_x = pos_json.get('ltX', 0)
-                            _f_x2 = pos_json.get('rbX', 0)
-                            _f_y = pos_json.get('ltY', 0)
-                            _f_y2 = pos_json.get('rbY', 0)
-                            _v_w = pos_json.get('imgWidth', 1)
-                            _v_h = pos_json.get('imgHeight', 1)
-                            if _v_w > 0 and _v_h > 0:
-                                # 计算坐标系统中的中心点
-                                center_x_ratio = (_f_x + _f_x2) / (2 * _v_w)
-                                center_y_ratio = (_f_y + _f_y2) / (2 * _v_h)
-                                # 转换为视频坐标系统
-                                center_x = f"iw*{center_x_ratio:.6f}"
-                                center_y = f"ih*{center_y_ratio:.6f}"
-                    except Exception as e:
-                        # 解析失败使用默认中心
-                        pass
-                    
-                    if duration == 0:
-                        # 静态缩放（整个视频时长）
-                        x_expr = f"({center_x})-(ow*zoom)/2"
-                        y_expr = f"({center_y})-(oh*zoom)/2"
-                        filter_args.append(f"{video_output_str}trim=start={start_time},zoompan=z={zoom_factor}:x={x_expr}:y={y_expr}:d=1[v_eff{effect_index}]")
-                    else:
-                        # 动态缩放（指定时间段内）
-                        zoom_expr = f"if(between(t\\,{start_time}\\,{start_time + duration})\\,{zoom_factor}\\,1)"
-                        x_expr = f"({center_x})-(ow*zoom)/2"
-                        y_expr = f"({center_y})-(oh*zoom)/2"
-                        filter_args.append(f"{video_output_str}zoompan=z={zoom_expr}:x={x_expr}:y={y_expr}:d=1[v_eff{effect_index}]")
-                    video_output_str = f"[v_eff{effect_index}]"
-                elif effect.startswith("skip:"):
-                    param = effect.split(":", 2)[1]
-                    if param == '':
-                        param = "0"
-                    skip_seconds = float(param)
-                    if skip_seconds > 0:
-                        effect_index += 1
-                        filter_args.append(f"{video_output_str}trim=start={skip_seconds}[v_eff{effect_index}]")
-                        video_output_str = f"[v_eff{effect_index}]"
-                elif effect.startswith("tail:"):
-                    param = effect.split(":", 2)[1]
-                    if param == '':
-                        param = "0"
-                    tail_seconds = float(param)
-                    if tail_seconds > 0:
-                        effect_index += 1
-                        # 首先获取视频总时长，然后计算开始时间
-                        # 使用reverse+trim+reverse的方法来精确获取最后N秒
-                        filter_args.append(f"{video_output_str}reverse[v_rev{effect_index}]")
-                        filter_args.append(f"[v_rev{effect_index}]trim=duration={tail_seconds}[v_trim{effect_index}]")
-                        filter_args.append(f"[v_trim{effect_index}]reverse[v_eff{effect_index}]")
-                        video_output_str = f"[v_eff{effect_index}]"
-                ...
-            if self.resolution:
-                filter_args.append(f"{video_output_str}scale={self.resolution.replace('x', ':')}[v]")
-                video_output_str = "[v]"
-            for lut in self.luts:
-                filter_args.append(f"{video_output_str}lut3d=file={lut}{video_output_str}")
-            for overlay in self.overlays:
-                input_index = input_args.count("-i")
-                input_args.append("-i")
-                input_args.append(overlay)
-                if os.getenv("OLD_FFMPEG"):
-                    filter_args.append(f"{video_output_str}[{input_index}:v]scale2ref=iw:ih[v]")
-                else:
-                    filter_args.append(f"{video_output_str}[{input_index}:v]scale=rw:rh[v]")
-                filter_args.append(f"[v][{input_index}:v]overlay=1:eof_action=endall[v]")
-                video_output_str = "[v]"
-            for subtitle in self.subtitles:
-                filter_args.append(f"{video_output_str}ass={subtitle}[v]")
-                video_output_str = "[v]"
-            output_args.append("-map")
-            output_args.append(video_output_str)
-            output_args.append("-r")
-            output_args.append(f"{self.frame_rate}")
-            output_args.append("-fps_mode")
-            output_args.append("cfr")
-            if self.mute:
-                input_index = input_args.count("-i")
-                input_args += MUTE_AUDIO_INPUT
-                filter_args.append(f"[{input_index}:a]acopy[a]")
-                audio_track_index += 1
-                audio_output_str = "[a]"
-            else:
-                audio_output_str = "[0:a]"
-                audio_track_index += 1
-            for audio in self.audios:
-                input_index = input_args.count("-i")
-                input_args.append("-i")
-                input_args.append(audio.replace("\\", "/"))
-                audio_track_index += 1
-                filter_args.append(f"{audio_output_str}[{input_index}:a]amix=duration=shortest:dropout_transition=0:normalize=0[a]")
-                audio_output_str = "[a]"
-            if audio_output_str:
-                output_args.append("-map")
-                output_args.append(audio_output_str)
-            _filter_args = [] if len(filter_args) == 0 else ["-filter_complex", ";".join(filter_args)]
-            return args + input_args + _filter_args + output_args + [self.get_output_file()]
-        elif self.task_type == 'concat':
-            # 无法通过 annexb 合并的
-            input_args = []
-            output_args = [*DEFAULT_ARGS]
-            filter_args = []
-            audio_output_str = ""
-            audio_track_index = 0
-            # output_args
-            if len(self.input_file) == 1:
-                _file = self.input_file[0]
-                from util.ffmpeg import probe_video_audio
-                if type(_file) is str:
-                    input_args += ["-i", _file]
-                    self.mute = not probe_video_audio(_file)
-                elif isinstance(_file, FfmpegTask):
-                    input_args += ["-i", _file.get_output_file()]
-                    self.mute = not probe_video_audio(_file.get_output_file())
-            else:
-                _tmp_file = "tmp_concat_" + str(time.time()) + ".txt"
-                from util.ffmpeg import probe_video_audio
-                with open(_tmp_file, "w", encoding="utf-8") as f:
-                    for input_file in self.input_file:
-                        if type(input_file) is str:
-                            f.write("file '" + input_file + "'\n")
-                        elif isinstance(input_file, FfmpegTask):
-                            f.write("file '" + input_file.get_output_file() + "'\n")
-                input_args += ["-f", "concat", "-safe", "0", "-i", _tmp_file]
-                self.mute = not probe_video_audio(_tmp_file, "concat")
-            output_args.append("-map")
-            output_args.append("0:v")
-            output_args.append("-c:v")
-            output_args.append("copy")
-            if self.mute:
-                input_index = input_args.count("-i")
-                input_args += MUTE_AUDIO_INPUT
-                audio_output_str = f"[{input_index}:a]"
-                audio_track_index += 1
-            else:
-                audio_output_str = "[0:a]"
-                audio_track_index += 1
-            for audio in self.audios:
-                input_index = input_args.count("-i")
-                input_args.append("-i")
-                input_args.append(audio.replace("\\", "/"))
-                audio_track_index += 1
-                filter_args.append(f"{audio_output_str}[{input_index}:a]amix=duration=shortest:dropout_transition=0:normalize=0[a]")
-                audio_output_str = "[a]"
-            if audio_output_str:
-                output_args.append("-map")
-                if audio_track_index <= 1:
-                    output_args.append(audio_output_str[1:-1])
-                else:
-                    output_args.append(audio_output_str)
-            output_args += AUDIO_ARGS
-            if self.annexb:
-                output_args.append("-bsf:v")
-                output_args.append(get_mp4toannexb_filter())
-                output_args.append("-bsf:a")
-                output_args.append("setts=pts=DTS")
-            output_args.append("-f")
-            output_args.append("mpegts" if self.annexb else "mp4")
-            _filter_args = [] if len(filter_args) == 0 else ["-filter_complex", ";".join(filter_args)]
-            return args + input_args + _filter_args + output_args + [self.get_output_file()]
-        elif self.task_type == 'copy':
-            if len(self.input_file) == 1:
-                if type(self.input_file[0]) is str:
-                    if self.input_file[0] == self.get_output_file():
-                        return []
-            return args + ["-i", self.input_file[0]] + ["-c", "copy", self.get_output_file()]
-        return []
-
-    def set_output_file(self, file=None):
-        if file is None:
-            if self.output_file == '':
-                if self.annexb:
-                    self.output_file = "rand_" + str(uuid.uuid4()) + ".ts"
-                else:
-                    self.output_file = "rand_" + str(uuid.uuid4()) + ".mp4"
-        else:
-            if isinstance(file, FfmpegTask):
-                if file == self:
-                    return
-                self.output_file = file.get_output_file()
-            if type(file) is str:
-                self.output_file = file
-
-    def check_annexb(self):
-        for input_file in self.input_file:
-            if type(input_file) is str:
-                if self.task_type == 'encode':
-                    return self.annexb
-                elif self.task_type == 'concat':
-                    return False
-                elif self.task_type == 'copy':
-                    return self.annexb
-                else:
-                    return False
-            elif isinstance(input_file, FfmpegTask):
-                if not input_file.check_annexb():
-                    return False
-        return True

handlers/__init__.py

@@ -0,0 +1,22 @@
+# -*- coding: utf-8 -*-
+"""
+任务处理器层
+
+包含各种任务类型的具体处理器实现。
+"""
+
+from handlers.base import BaseHandler
+from handlers.render_video import RenderSegmentVideoHandler
+from handlers.compose_transition import ComposeTransitionHandler
+from handlers.prepare_audio import PrepareJobAudioHandler
+from handlers.package_ts import PackageSegmentTsHandler
+from handlers.finalize_mp4 import FinalizeMp4Handler
+
+__all__ = [
+    'BaseHandler',
+    'RenderSegmentVideoHandler',
+    'ComposeTransitionHandler',
+    'PrepareJobAudioHandler',
+    'PackageSegmentTsHandler',
+    'FinalizeMp4Handler',
+]

handlers/base.py

@@ -0,0 +1,580 @@
+# -*- coding: utf-8 -*-
+"""
+任务处理器基类
+
+提供所有处理器共用的基础功能。
+"""
+
+import os
+import json
+import logging
+import shutil
+import tempfile
+import subprocess
+import threading
+from abc import ABC
+from typing import Optional, List, Dict, Any, Tuple, TYPE_CHECKING
+
+from core.handler import TaskHandler
+from domain.task import Task
+from domain.result import TaskResult, ErrorCode
+from domain.config import WorkerConfig
+from services import storage
+from services.cache import MaterialCache
+from constant import (
+    HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA,
+    VIDEO_ENCODE_PARAMS, VIDEO_ENCODE_PARAMS_QSV, VIDEO_ENCODE_PARAMS_CUDA
+)
+
+if TYPE_CHECKING:
+    from services.api_client import APIClientV2
+
+logger = logging.getLogger(__name__)
+
+
+def get_video_encode_args(hw_accel: str = HW_ACCEL_NONE) -> List[str]:
+    """
+    根据硬件加速配置获取视频编码参数
+
+    Args:
+        hw_accel: 硬件加速类型 (none, qsv, cuda)
+
+    Returns:
+        FFmpeg 视频编码参数列表
+    """
+    if hw_accel == HW_ACCEL_QSV:
+        params = VIDEO_ENCODE_PARAMS_QSV
+        return [
+            '-c:v', params['codec'],
+            '-preset', params['preset'],
+            '-profile:v', params['profile'],
+            '-level', params['level'],
+            '-global_quality', params['global_quality'],
+            '-look_ahead', params['look_ahead'],
+        ]
+    elif hw_accel == HW_ACCEL_CUDA:
+        params = VIDEO_ENCODE_PARAMS_CUDA
+        return [
+            '-c:v', params['codec'],
+            '-preset', params['preset'],
+            '-profile:v', params['profile'],
+            '-level', params['level'],
+            '-rc', params['rc'],
+            '-cq', params['cq'],
+            '-b:v', '0',  # 配合 vbr 模式使用 cq
+        ]
+    else:
+        # 软件编码（默认）
+        params = VIDEO_ENCODE_PARAMS
+        return [
+            '-c:v', params['codec'],
+            '-preset', params['preset'],
+            '-profile:v', params['profile'],
+            '-level', params['level'],
+            '-crf', params['crf'],
+            '-pix_fmt', params['pix_fmt'],
+        ]
+
+
+def get_hwaccel_decode_args(hw_accel: str = HW_ACCEL_NONE, device_index: Optional[int] = None) -> List[str]:
+    """
+    获取硬件加速解码参数（输入文件之前使用）
+
+    Args:
+        hw_accel: 硬件加速类型 (none, qsv, cuda)
+        device_index: GPU 设备索引，用于多显卡调度
+
+    Returns:
+        FFmpeg 硬件加速解码参数列表
+    """
+    if hw_accel == HW_ACCEL_CUDA:
+        # CUDA 硬件加速解码
+        args = ['-hwaccel', 'cuda']
+        # 多显卡模式下指定设备
+        if device_index is not None:
+            args.extend(['-hwaccel_device', str(device_index)])
+        args.extend(['-hwaccel_output_format', 'cuda'])
+        return args
+    elif hw_accel == HW_ACCEL_QSV:
+        # QSV 硬件加速解码
+        args = ['-hwaccel', 'qsv']
+        # QSV 在 Windows 上使用 -qsv_device
+        if device_index is not None:
+            args.extend(['-qsv_device', str(device_index)])
+        args.extend(['-hwaccel_output_format', 'qsv'])
+        return args
+    else:
+        return []
+
+
+def get_hwaccel_filter_prefix(hw_accel: str = HW_ACCEL_NONE) -> str:
+    """
+    获取硬件加速滤镜前缀（用于 hwdownload 从 GPU 到 CPU）
+
+    注意：由于大多数复杂滤镜（如 lut3d, overlay, crop 等）不支持硬件表面，
+    我们需要在滤镜链开始时将硬件表面下载到系统内存。
+
+    CUDA/QSV hwdownload 只支持 nv12 格式输出，因此需要两步转换：
+    1. hwdownload,format=nv12 - 从 GPU 下载到 CPU
+    2. format=yuv420p - 转换为标准格式（确保与 RGBA/YUVA overlay 混合时颜色正确）
+
+    Args:
+        hw_accel: 硬件加速类型
+
+    Returns:
+        需要添加到滤镜链开头的 hwdownload 滤镜字符串
+    """
+    if hw_accel == HW_ACCEL_CUDA:
+        return 'hwdownload,format=nv12,format=yuv420p,'
+    elif hw_accel == HW_ACCEL_QSV:
+        return 'hwdownload,format=nv12,format=yuv420p,'
+    else:
+        return ''
+
+
+# v2 统一视频编码参数（兼容旧代码，使用软件编码）
+VIDEO_ENCODE_ARGS = get_video_encode_args(HW_ACCEL_NONE)
+
+# v2 统一音频编码参数
+AUDIO_ENCODE_ARGS = [
+    '-c:a', 'aac',
+    '-b:a', '128k',
+    '-ar', '48000',
+    '-ac', '2',
+]
+
+FFMPEG_LOGLEVEL = 'error'
+
+
+def subprocess_args(include_stdout: bool = True) -> Dict[str, Any]:
+    """
+    创建跨平台的 subprocess 参数
+
+    在 Windows 上使用 Pyinstaller --noconsole 打包时，需要特殊处理以避免弹出命令行窗口。
+
+    Args:
+        include_stdout: 是否包含 stdout 捕获
+
+    Returns:
+        subprocess.run 使用的参数字典
+    """
+    ret: Dict[str, Any] = {}
+
+    # Windows 特殊处理
+    if hasattr(subprocess, 'STARTUPINFO'):
+        si = subprocess.STARTUPINFO()
+        si.dwFlags |= subprocess.STARTF_USESHOWWINDOW
+        ret['startupinfo'] = si
+        ret['env'] = os.environ
+
+    # 重定向 stdin 避免 "handle is invalid" 错误
+    ret['stdin'] = subprocess.PIPE
+
+    if include_stdout:
+        ret['stdout'] = subprocess.PIPE
+
+    return ret
+
+
+def probe_video_info(video_file: str) -> Tuple[int, int, float]:
+    """
+    探测视频信息（宽度、高度、时长）
+
+    Args:
+        video_file: 视频文件路径
+
+    Returns:
+        (width, height, duration) 元组，失败返回 (0, 0, 0)
+    """
+    try:
+        result = subprocess.run(
+            [
+                'ffprobe', '-v', 'error',
+                '-select_streams', 'v:0',
+                '-show_entries', 'stream=width,height:format=duration',
+                '-of', 'csv=s=x:p=0',
+                video_file
+            ],
+            capture_output=True,
+            timeout=30,
+            **subprocess_args(False)
+        )
+
+        if result.returncode != 0:
+            logger.warning(f"ffprobe failed for {video_file}")
+            return 0, 0, 0
+
+        output = result.stdout.decode('utf-8').strip()
+        if not output:
+            return 0, 0, 0
+
+        lines = output.split('\n')
+        if len(lines) >= 2:
+            wh = lines[0].strip()
+            duration_str = lines[1].strip()
+            width, height = wh.split('x')
+            return int(width), int(height), float(duration_str)
+
+        return 0, 0, 0
+
+    except Exception as e:
+        logger.warning(f"probe_video_info error: {e}")
+        return 0, 0, 0
+
+
+def probe_duration_json(file_path: str) -> Optional[float]:
+    """
+    使用 ffprobe JSON 输出探测媒体时长
+
+    Args:
+        file_path: 媒体文件路径
+
+    Returns:
+        时长（秒），失败返回 None
+    """
+    try:
+        result = subprocess.run(
+            [
+                'ffprobe', '-v', 'error',
+                '-show_entries', 'format=duration',
+                '-of', 'json',
+                file_path
+            ],
+            capture_output=True,
+            timeout=30,
+            **subprocess_args(False)
+        )
+
+        if result.returncode != 0:
+            return None
+
+        data = json.loads(result.stdout.decode('utf-8'))
+        duration = data.get('format', {}).get('duration')
+        return float(duration) if duration else None
+
+    except Exception as e:
+        logger.warning(f"probe_duration_json error: {e}")
+        return None
+
+
+class BaseHandler(TaskHandler, ABC):
+    """
+    任务处理器基类
+
+    提供所有处理器共用的基础功能，包括：
+    - 临时目录管理
+    - 文件下载/上传
+    - FFmpeg 命令执行
+    - GPU 设备管理（多显卡调度）
+    - 日志记录
+    """
+
+    # 线程本地存储：用于存储当前线程的 GPU 设备索引
+    _thread_local = threading.local()
+
+    def __init__(self, config: WorkerConfig, api_client: 'APIClientV2'):
+        """
+        初始化处理器
+
+        Args:
+            config: Worker 配置
+            api_client: API 客户端
+        """
+        self.config = config
+        self.api_client = api_client
+        self.material_cache = MaterialCache(
+            cache_dir=config.cache_dir,
+            enabled=config.cache_enabled,
+            max_size_gb=config.cache_max_size_gb
+        )
+
+    # ========== GPU 设备管理 ==========
+
+    def set_gpu_device(self, device_index: int) -> None:
+        """
+        设置当前线程的 GPU 设备索引
+
+        由 TaskExecutor 在任务执行前调用。
+
+        Args:
+            device_index: GPU 设备索引
+        """
+        self._thread_local.gpu_device = device_index
+
+    def get_gpu_device(self) -> Optional[int]:
+        """
+        获取当前线程的 GPU 设备索引
+
+        Returns:
+            GPU 设备索引，未设置则返回 None
+        """
+        return getattr(self._thread_local, 'gpu_device', None)
+
+    def clear_gpu_device(self) -> None:
+        """
+        清除当前线程的 GPU 设备索引
+
+        由 TaskExecutor 在任务执行后调用。
+        """
+        if hasattr(self._thread_local, 'gpu_device'):
+            del self._thread_local.gpu_device
+
+    # ========== FFmpeg 参数生成 ==========
+
+    def get_video_encode_args(self) -> List[str]:
+        """
+        获取当前配置的视频编码参数
+
+        Returns:
+            FFmpeg 视频编码参数列表
+        """
+        return get_video_encode_args(self.config.hw_accel)
+
+    def get_hwaccel_decode_args(self) -> List[str]:
+        """
+        获取硬件加速解码参数（支持设备指定）
+
+        Returns:
+            FFmpeg 硬件加速解码参数列表
+        """
+        device_index = self.get_gpu_device()
+        return get_hwaccel_decode_args(self.config.hw_accel, device_index)
+
+    def get_hwaccel_filter_prefix(self) -> str:
+        """
+        获取硬件加速滤镜前缀
+
+        Returns:
+            需要添加到滤镜链开头的 hwdownload 滤镜字符串
+        """
+        return get_hwaccel_filter_prefix(self.config.hw_accel)
+
+    def before_handle(self, task: Task) -> None:
+        """处理前钩子"""
+        logger.debug(f"[task:{task.task_id}] Before handle: {task.task_type.value}")
+
+    def after_handle(self, task: Task, result: TaskResult) -> None:
+        """处理后钩子"""
+        status = "success" if result.success else "failed"
+        logger.debug(f"[task:{task.task_id}] After handle: {status}")
+
+    def create_work_dir(self, task_id: str = None) -> str:
+        """
+        创建临时工作目录
+
+        Args:
+            task_id: 任务 ID（用于目录命名）
+
+        Returns:
+            工作目录路径
+        """
+        # 确保临时根目录存在
+        os.makedirs(self.config.temp_dir, exist_ok=True)
+
+        # 创建唯一的工作目录
+        prefix = f"task_{task_id}_" if task_id else "task_"
+        work_dir = tempfile.mkdtemp(dir=self.config.temp_dir, prefix=prefix)
+
+        logger.debug(f"Created work directory: {work_dir}")
+        return work_dir
+
+    def cleanup_work_dir(self, work_dir: str) -> None:
+        """
+        清理临时工作目录
+
+        Args:
+            work_dir: 工作目录路径
+        """
+        if not work_dir or not os.path.exists(work_dir):
+            return
+
+        try:
+            shutil.rmtree(work_dir)
+            logger.debug(f"Cleaned up work directory: {work_dir}")
+        except Exception as e:
+            logger.warning(f"Failed to cleanup work directory {work_dir}: {e}")
+
+    def download_file(self, url: str, dest: str, timeout: int = None, use_cache: bool = True) -> bool:
+        """
+        下载文件（支持缓存）
+
+        Args:
+            url: 文件 URL
+            dest: 目标路径
+            timeout: 超时时间（秒）
+            use_cache: 是否使用缓存（默认 True）
+
+        Returns:
+            是否成功
+        """
+        if timeout is None:
+            timeout = self.config.download_timeout
+
+        try:
+            if use_cache:
+                # 使用缓存下载
+                result = self.material_cache.get_or_download(url, dest, timeout=timeout)
+            else:
+                # 直接下载（不走缓存）
+                result = storage.download_file(url, dest, timeout=timeout)
+
+            if result:
+                file_size = os.path.getsize(dest) if os.path.exists(dest) else 0
+                logger.debug(f"Downloaded: {url} -> {dest} ({file_size} bytes)")
+            return result
+        except Exception as e:
+            logger.error(f"Download failed: {url} -> {e}")
+            return False
+
+    def upload_file(
+        self,
+        task_id: str,
+        file_type: str,
+        file_path: str,
+        file_name: str = None
+    ) -> Optional[str]:
+        """
+        上传文件并返回访问 URL
+
+        Args:
+            task_id: 任务 ID
+            file_type: 文件类型（video/audio/ts/mp4）
+            file_path: 本地文件路径
+            file_name: 文件名（可选）
+
+        Returns:
+            访问 URL，失败返回 None
+        """
+        # 获取上传 URL
+        upload_info = self.api_client.get_upload_url(task_id, file_type, file_name)
+        if not upload_info:
+            logger.error(f"[task:{task_id}] Failed to get upload URL")
+            return None
+
+        upload_url = upload_info.get('uploadUrl')
+        access_url = upload_info.get('accessUrl')
+
+        if not upload_url:
+            logger.error(f"[task:{task_id}] Invalid upload URL response")
+            return None
+
+        # 上传文件
+        try:
+            result = storage.upload_file(upload_url, file_path, timeout=self.config.upload_timeout)
+            if result:
+                file_size = os.path.getsize(file_path)
+                logger.info(f"[task:{task_id}] Uploaded: {file_path} ({file_size} bytes)")
+                return access_url
+            else:
+                logger.error(f"[task:{task_id}] Upload failed: {file_path}")
+                return None
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Upload error: {e}")
+            return None
+
+    def run_ffmpeg(
+        self,
+        cmd: List[str],
+        task_id: str,
+        timeout: int = None
+    ) -> bool:
+        """
+        执行 FFmpeg 命令
+
+        Args:
+            cmd: FFmpeg 命令参数列表
+            task_id: 任务 ID（用于日志）
+            timeout: 超时时间（秒）
+
+        Returns:
+            是否成功
+        """
+        if timeout is None:
+            timeout = self.config.ffmpeg_timeout
+
+        cmd_to_run = list(cmd)
+        if cmd_to_run and cmd_to_run[0] == 'ffmpeg' and '-loglevel' not in cmd_to_run:
+            cmd_to_run[1:1] = ['-loglevel', FFMPEG_LOGLEVEL]
+
+        # 日志记录命令（限制长度）
+        cmd_str = ' '.join(cmd_to_run)
+        if len(cmd_str) > 500:
+            cmd_str = cmd_str[:500] + '...'
+        logger.info(f"[task:{task_id}] FFmpeg: {cmd_str}")
+
+        try:
+            run_args = subprocess_args(False)
+            run_args['stdout'] = subprocess.DEVNULL
+            run_args['stderr'] = subprocess.PIPE
+            result = subprocess.run(
+                cmd_to_run,
+                timeout=timeout,
+                **run_args
+            )
+
+            if result.returncode != 0:
+                stderr = (result.stderr or b'').decode('utf-8', errors='replace')[:1000]
+                logger.error(f"[task:{task_id}] FFmpeg failed (code={result.returncode}): {stderr}")
+                return False
+
+            return True
+
+        except subprocess.TimeoutExpired:
+            logger.error(f"[task:{task_id}] FFmpeg timeout after {timeout}s")
+            return False
+        except Exception as e:
+            logger.error(f"[task:{task_id}] FFmpeg error: {e}")
+            return False
+
+    def probe_duration(self, file_path: str) -> Optional[float]:
+        """
+        探测媒体文件时长
+
+        Args:
+            file_path: 文件路径
+
+        Returns:
+            时长（秒），失败返回 None
+        """
+        # 首先尝试 JSON 输出方式
+        duration = probe_duration_json(file_path)
+        if duration is not None:
+            return duration
+
+        # 回退到旧方式
+        try:
+            _, _, duration = probe_video_info(file_path)
+            return float(duration) if duration else None
+        except Exception as e:
+            logger.warning(f"Failed to probe duration: {file_path} -> {e}")
+            return None
+
+    def get_file_size(self, file_path: str) -> int:
+        """
+        获取文件大小
+
+        Args:
+            file_path: 文件路径
+
+        Returns:
+            文件大小（字节）
+        """
+        try:
+            return os.path.getsize(file_path)
+        except Exception:
+            return 0
+
+    def ensure_file_exists(self, file_path: str, min_size: int = 0) -> bool:
+        """
+        确保文件存在且大小满足要求
+
+        Args:
+            file_path: 文件路径
+            min_size: 最小大小（字节）
+
+        Returns:
+            是否满足要求
+        """
+        if not os.path.exists(file_path):
+            return False
+        return os.path.getsize(file_path) >= min_size

handlers/compose_transition.py

@@ -0,0 +1,273 @@
+# -*- coding: utf-8 -*-
+"""
+转场合成处理器
+
+处理 COMPOSE_TRANSITION 任务，将相邻两个片段的 overlap 区域进行混合，生成转场效果。
+使用 FFmpeg xfade 滤镜实现多种转场效果。
+"""
+
+import os
+import logging
+from typing import List, Optional
+
+from handlers.base import BaseHandler
+from domain.task import Task, TaskType, TransitionConfig, TRANSITION_TYPES
+from domain.result import TaskResult, ErrorCode
+
+logger = logging.getLogger(__name__)
+
+
+class ComposeTransitionHandler(BaseHandler):
+    """
+    转场合成处理器
+
+    职责：
+    - 下载前一个片段的视频（含尾部 overlap）
+    - 下载后一个片段的视频（含头部 overlap）
+    - 使用 xfade 滤镜合成转场效果
+    - 上传转场视频产物
+
+    关键约束：
+    - 转场任务必须等待前后两个片段的 RENDER_SEGMENT_VIDEO 都完成后才能执行
+    - 输出编码参数必须与片段视频一致，确保后续 TS 封装兼容
+    - 转场视频不含音频轨道（音频由 PREPARE_JOB_AUDIO 统一处理）
+    """
+
+    def get_supported_type(self) -> TaskType:
+        return TaskType.COMPOSE_TRANSITION
+
+    def handle(self, task: Task) -> TaskResult:
+        """处理转场合成任务"""
+        work_dir = self.create_work_dir(task.task_id)
+
+        try:
+            # 解析参数
+            transition_id = task.get_transition_id()
+            prev_segment = task.get_prev_segment()
+            next_segment = task.get_next_segment()
+            transition_config = task.get_transition_config()
+            output_spec = task.get_output_spec()
+
+            # 参数验证
+            if not transition_id:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing transitionId"
+                )
+
+            if not prev_segment or not prev_segment.get('videoUrl'):
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing prevSegment.videoUrl"
+                )
+
+            if not next_segment or not next_segment.get('videoUrl'):
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing nextSegment.videoUrl"
+                )
+
+            if not transition_config:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing transition config"
+                )
+
+            # 获取 overlap 时长
+            overlap_tail_ms = prev_segment.get('overlapTailMs', 0)
+            overlap_head_ms = next_segment.get('overlapHeadMs', 0)
+            transition_duration_ms = transition_config.duration_ms
+
+            # 验证 overlap 时长
+            if overlap_tail_ms <= 0 or overlap_head_ms <= 0:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    f"Invalid overlap duration: tail={overlap_tail_ms}ms, head={overlap_head_ms}ms"
+                )
+
+            logger.info(
+                f"[task:{task.task_id}] Composing transition: {transition_config.type}, "
+                f"duration={transition_duration_ms}ms, "
+                f"overlap_tail={overlap_tail_ms}ms, overlap_head={overlap_head_ms}ms"
+            )
+
+            # 1. 下载前一个片段视频
+            prev_video_file = os.path.join(work_dir, 'prev_segment.mp4')
+            if not self.download_file(prev_segment['videoUrl'], prev_video_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download prev segment video: {prev_segment['videoUrl']}"
+                )
+
+            # 2. 下载后一个片段视频
+            next_video_file = os.path.join(work_dir, 'next_segment.mp4')
+            if not self.download_file(next_segment['videoUrl'], next_video_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download next segment video: {next_segment['videoUrl']}"
+                )
+
+            # 3. 获取前一个片段的实际时长
+            prev_duration = self.probe_duration(prev_video_file)
+            if not prev_duration:
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "Failed to probe prev segment duration"
+                )
+
+            # 4. 构建转场合成命令
+            output_file = os.path.join(work_dir, 'transition.mp4')
+            cmd = self._build_command(
+                prev_video_file=prev_video_file,
+                next_video_file=next_video_file,
+                output_file=output_file,
+                prev_duration_sec=prev_duration,
+                overlap_tail_ms=overlap_tail_ms,
+                overlap_head_ms=overlap_head_ms,
+                transition_config=transition_config,
+                output_spec=output_spec
+            )
+
+            # 5. 执行 FFmpeg
+            if not self.run_ffmpeg(cmd, task.task_id):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "FFmpeg transition composition failed"
+                )
+
+            # 6. 验证输出文件
+            if not self.ensure_file_exists(output_file, min_size=1024):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "Transition output file is missing or too small"
+                )
+
+            # 7. 获取实际时长
+            actual_duration = self.probe_duration(output_file)
+            actual_duration_ms = int(actual_duration * 1000) if actual_duration else transition_duration_ms
+
+            # 8. 上传产物
+            transition_video_url = self.upload_file(task.task_id, 'video', output_file)
+            if not transition_video_url:
+                return TaskResult.fail(
+                    ErrorCode.E_UPLOAD_FAILED,
+                    "Failed to upload transition video"
+                )
+
+            return TaskResult.ok({
+                'transitionVideoUrl': transition_video_url,
+                'actualDurationMs': actual_duration_ms
+            })
+
+        except Exception as e:
+            logger.error(f"[task:{task.task_id}] Unexpected error: {e}", exc_info=True)
+            return TaskResult.fail(ErrorCode.E_UNKNOWN, str(e))
+
+        finally:
+            self.cleanup_work_dir(work_dir)
+
+    def _build_command(
+        self,
+        prev_video_file: str,
+        next_video_file: str,
+        output_file: str,
+        prev_duration_sec: float,
+        overlap_tail_ms: int,
+        overlap_head_ms: int,
+        transition_config: TransitionConfig,
+        output_spec
+    ) -> List[str]:
+        """
+        构建转场合成命令
+
+        使用 xfade 滤镜合成转场效果：
+        1. 从前一个片段截取尾部 overlap 区域
+        2. 从后一个片段截取头部 overlap 区域
+        3. 使用 xfade 进行混合
+
+        注意：
+        - 转场视频时长很短，需要特别处理 GOP 大小
+        - 确保第一帧是关键帧以便后续 TS 封装
+
+        Args:
+            prev_video_file: 前一个片段视频路径
+            next_video_file: 后一个片段视频路径
+            output_file: 输出文件路径
+            prev_duration_sec: 前一个片段总时长（秒）
+            overlap_tail_ms: 尾部 overlap 时长（毫秒）
+            overlap_head_ms: 头部 overlap 时长（毫秒）
+            transition_config: 转场配置
+            output_spec: 输出规格
+
+        Returns:
+            FFmpeg 命令参数列表
+        """
+        # 计算时间参数
+        overlap_tail_sec = overlap_tail_ms / 1000.0
+        overlap_head_sec = overlap_head_ms / 1000.0
+
+        # 前一个片段的尾部 overlap 起始位置
+        tail_start_sec = prev_duration_sec - overlap_tail_sec
+
+        # 转场时长（使用两个 overlap 区域的总和，xfade 会将两段合成为此时长）
+        # 注意：xfade 的输出时长 = overlap_tail + overlap_head - duration
+        # 当 duration = overlap_tail + overlap_head 时，输出时长约等于 duration
+        transition_duration_sec = min(overlap_tail_sec, overlap_head_sec)
+
+        # 获取 xfade 转场类型
+        xfade_transition = transition_config.get_ffmpeg_transition()
+
+        # 构建滤镜
+        # [0:v] trim 截取前一个片段的尾部 overlap
+        # [1:v] trim 截取后一个片段的头部 overlap
+        # xfade 混合两段视频
+        filter_complex = (
+            f"[0:v]trim=start={tail_start_sec},setpts=PTS-STARTPTS[v0];"
+            f"[1:v]trim=end={overlap_head_sec},setpts=PTS-STARTPTS[v1];"
+            f"[v0][v1]xfade=transition={xfade_transition}:duration={transition_duration_sec}:offset=0[outv]"
+        )
+
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            '-i', prev_video_file,
+            '-i', next_video_file,
+            '-filter_complex', filter_complex,
+            '-map', '[outv]',
+        ]
+
+        # 编码参数（根据硬件加速配置动态获取）
+        cmd.extend(self.get_video_encode_args())
+
+        # 帧率
+        fps = output_spec.fps
+
+        # 计算输出视频的预估帧数
+        # xfade 输出时长 ≈ overlap_tail + overlap_head - transition_duration
+        output_duration_sec = overlap_tail_sec + overlap_head_sec - transition_duration_sec
+        total_frames = int(output_duration_sec * fps)
+
+        # 动态调整 GOP 大小：对于短视频，GOP 不能大于总帧数
+        # 确保至少有 1 个关键帧（第一帧），最小 GOP = 1
+        if total_frames <= 1:
+            gop_size = 1
+        elif total_frames < fps:
+            # 短于 1 秒的视频，使用全部帧数作为 GOP（整个视频只有开头一个关键帧）
+            gop_size = total_frames
+        else:
+            # 正常情况，每秒一个关键帧（比标准的 2 秒更密集，适合短视频）
+            gop_size = fps
+
+        cmd.extend(['-r', str(fps)])
+        cmd.extend(['-g', str(gop_size)])
+        cmd.extend(['-keyint_min', str(min(gop_size, fps // 2 or 1))])
+
+        # 强制第一帧为关键帧
+        cmd.extend(['-force_key_frames', 'expr:eq(n,0)'])
+
+        # 无音频
+        cmd.append('-an')
+
+        # 输出文件
+        cmd.append(output_file)
+
+        return cmd

handlers/finalize_mp4.py

@@ -0,0 +1,190 @@
+# -*- coding: utf-8 -*-
+"""
+最终 MP4 合并处理器
+
+处理 FINALIZE_MP4 任务，将所有 TS 分片合并为最终可下载的 MP4 文件。
+"""
+
+import os
+import logging
+from typing import List
+
+from handlers.base import BaseHandler
+from domain.task import Task, TaskType
+from domain.result import TaskResult, ErrorCode
+
+logger = logging.getLogger(__name__)
+
+
+class FinalizeMp4Handler(BaseHandler):
+    """
+    最终 MP4 合并处理器
+
+    职责：
+    - 下载所有 TS 分片
+    - 使用 concat demuxer 合并
+    - 产出最终 MP4（remux，不重编码）
+    - 上传 MP4 产物
+
+    关键约束：
+    - 优先使用 remux（复制流，不重新编码）
+    - 使用 aac_adtstoasc bitstream filter 处理音频
+    """
+
+    def get_supported_type(self) -> TaskType:
+        return TaskType.FINALIZE_MP4
+
+    def handle(self, task: Task) -> TaskResult:
+        """处理 MP4 合并任务"""
+        work_dir = self.create_work_dir(task.task_id)
+
+        try:
+            # 获取 TS 列表
+            ts_list = task.get_ts_list()
+            m3u8_url = task.get_m3u8_url()
+
+            if not ts_list and not m3u8_url:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing tsList or m3u8Url"
+                )
+
+            output_file = os.path.join(work_dir, 'final.mp4')
+
+            if ts_list:
+                # 方式1：使用 TS 列表
+                result = self._process_ts_list(task, work_dir, ts_list, output_file)
+            else:
+                # 方式2：使用 m3u8 URL
+                result = self._process_m3u8(task, work_dir, m3u8_url, output_file)
+
+            if not result.success:
+                return result
+
+            # 验证输出文件
+            if not self.ensure_file_exists(output_file, min_size=4096):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "MP4 output file is missing or too small"
+                )
+
+            # 获取文件大小
+            file_size = self.get_file_size(output_file)
+
+            # 上传产物
+            mp4_url = self.upload_file(task.task_id, 'mp4', output_file)
+            if not mp4_url:
+                return TaskResult.fail(
+                    ErrorCode.E_UPLOAD_FAILED,
+                    "Failed to upload MP4"
+                )
+
+            return TaskResult.ok({
+                'mp4Url': mp4_url,
+                'fileSizeBytes': file_size
+            })
+
+        except Exception as e:
+            logger.error(f"[task:{task.task_id}] Unexpected error: {e}", exc_info=True)
+            return TaskResult.fail(ErrorCode.E_UNKNOWN, str(e))
+
+        finally:
+            self.cleanup_work_dir(work_dir)
+
+    def _process_ts_list(
+        self,
+        task: Task,
+        work_dir: str,
+        ts_list: List[str],
+        output_file: str
+    ) -> TaskResult:
+        """
+        使用 TS 列表处理
+
+        Args:
+            task: 任务实体
+            work_dir: 工作目录
+            ts_list: TS URL 列表
+            output_file: 输出文件路径
+
+        Returns:
+            TaskResult
+        """
+        # 1. 下载所有 TS 分片
+        ts_files = []
+        for i, ts_url in enumerate(ts_list):
+            ts_file = os.path.join(work_dir, f'seg_{i}.ts')
+            if not self.download_file(ts_url, ts_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download TS segment {i}: {ts_url}"
+                )
+            ts_files.append(ts_file)
+
+        logger.info(f"[task:{task.task_id}] Downloaded {len(ts_files)} TS segments")
+
+        # 2. 创建 concat 文件列表
+        concat_file = os.path.join(work_dir, 'concat.txt')
+        with open(concat_file, 'w', encoding='utf-8') as f:
+            for ts_file in ts_files:
+                # 路径中的反斜杠需要转义或使用正斜杠
+                ts_path = ts_file.replace('\\', '/')
+                f.write(f"file '{ts_path}'\n")
+
+        # 3. 构建合并命令（remux，不重编码）
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            '-f', 'concat',
+            '-safe', '0',
+            '-i', concat_file,
+            '-c', 'copy',  # 复制流，不重编码
+            '-bsf:a', 'aac_adtstoasc',  # 音频 bitstream filter
+            output_file
+        ]
+
+        # 4. 执行 FFmpeg
+        if not self.run_ffmpeg(cmd, task.task_id):
+            return TaskResult.fail(
+                ErrorCode.E_FFMPEG_FAILED,
+                "MP4 concatenation failed"
+            )
+
+        return TaskResult.ok({})
+
+    def _process_m3u8(
+        self,
+        task: Task,
+        work_dir: str,
+        m3u8_url: str,
+        output_file: str
+    ) -> TaskResult:
+        """
+        使用 m3u8 URL 处理
+
+        Args:
+            task: 任务实体
+            work_dir: 工作目录
+            m3u8_url: m3u8 URL
+            output_file: 输出文件路径
+
+        Returns:
+            TaskResult
+        """
+        # 构建命令
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            '-protocol_whitelist', 'file,http,https,tcp,tls',
+            '-i', m3u8_url,
+            '-c', 'copy',
+            '-bsf:a', 'aac_adtstoasc',
+            output_file
+        ]
+
+        # 执行 FFmpeg
+        if not self.run_ffmpeg(cmd, task.task_id):
+            return TaskResult.fail(
+                ErrorCode.E_FFMPEG_FAILED,
+                "MP4 conversion from m3u8 failed"
+            )
+
+        return TaskResult.ok({})

handlers/package_ts.py

@@ -0,0 +1,304 @@
+# -*- coding: utf-8 -*-
+"""
+TS 分片封装处理器
+
+处理 PACKAGE_SEGMENT_TS 任务，将视频片段和对应时间区间的音频封装为 TS 分片。
+支持转场相关的 overlap 裁剪和转场分片封装。
+"""
+
+import os
+import logging
+from typing import List, Optional
+
+from handlers.base import BaseHandler, VIDEO_ENCODE_ARGS
+from domain.task import Task, TaskType
+from domain.result import TaskResult, ErrorCode
+
+logger = logging.getLogger(__name__)
+
+
+class PackageSegmentTsHandler(BaseHandler):
+    """
+    TS 分片封装处理器
+
+    职责：
+    - 下载视频片段
+    - 下载全局音频
+    - 截取对应时间区间的音频
+    - 封装为 TS 分片
+    - 上传 TS 产物
+
+    关键约束：
+    - TS 必须包含音视频同轨
+    - 使用 output_ts_offset 保证时间戳连续
+    - 输出 extinfDurationSec 供 m3u8 使用
+
+    转场相关：
+    - 普通片段 TS：需要裁剪掉 overlap 区域（已被转场分片使用）
+    - 转场分片 TS：直接封装转场视频产物，无需裁剪
+    - 无转场时：走原有逻辑，不做裁剪
+
+    精确裁剪：
+    - 当需要裁剪 overlap 区域时，必须使用重编码方式（-vf trim）才能精确切割
+    - 使用 -c copy 只能从关键帧切割，会导致不精确
+    """
+
+    def get_supported_type(self) -> TaskType:
+        return TaskType.PACKAGE_SEGMENT_TS
+
+    def handle(self, task: Task) -> TaskResult:
+        """处理 TS 封装任务"""
+        work_dir = self.create_work_dir(task.task_id)
+
+        try:
+            # 解析参数
+            video_url = task.get_video_url()
+            audio_url = task.get_audio_url()
+            start_time_ms = task.get_start_time_ms()
+            duration_ms = task.get_duration_ms()
+            output_spec = task.get_output_spec()
+
+            # 转场相关参数
+            is_transition_segment = task.is_transition_segment()
+            trim_head = task.should_trim_head()
+            trim_tail = task.should_trim_tail()
+            trim_head_ms = task.get_trim_head_ms()
+            trim_tail_ms = task.get_trim_tail_ms()
+
+            if not video_url:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing videoUrl"
+                )
+
+            if not audio_url:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing audioUrl"
+                )
+
+            # 计算时间参数
+            start_sec = start_time_ms / 1000.0
+            duration_sec = duration_ms / 1000.0
+
+            # 1. 下载视频片段
+            video_file = os.path.join(work_dir, 'video.mp4')
+            if not self.download_file(video_url, video_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download video: {video_url}"
+                )
+
+            # 2. 下载全局音频
+            audio_file = os.path.join(work_dir, 'audio.aac')
+            if not self.download_file(audio_url, audio_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download audio: {audio_url}"
+                )
+
+            # 3. 判断是否需要精确裁剪视频
+            needs_video_trim = not is_transition_segment and (
+                (trim_head and trim_head_ms > 0) or
+                (trim_tail and trim_tail_ms > 0)
+            )
+
+            # 4. 如果需要裁剪，先重编码裁剪视频
+            processed_video_file = video_file
+            if needs_video_trim:
+                processed_video_file = os.path.join(work_dir, 'trimmed_video.mp4')
+                trim_cmd = self._build_trim_command(
+                    video_file=video_file,
+                    output_file=processed_video_file,
+                    trim_head_ms=trim_head_ms if trim_head else 0,
+                    trim_tail_ms=trim_tail_ms if trim_tail else 0,
+                    output_spec=output_spec
+                )
+
+                logger.info(f"[task:{task.task_id}] Trimming video: head={trim_head_ms}ms, tail={trim_tail_ms}ms")
+
+                if not self.run_ffmpeg(trim_cmd, task.task_id):
+                    return TaskResult.fail(
+                        ErrorCode.E_FFMPEG_FAILED,
+                        "Video trim failed"
+                    )
+
+                if not self.ensure_file_exists(processed_video_file, min_size=1024):
+                    return TaskResult.fail(
+                        ErrorCode.E_FFMPEG_FAILED,
+                        "Trimmed video file is missing or too small"
+                    )
+
+            # 5. 构建 TS 封装命令
+            output_file = os.path.join(work_dir, 'segment.ts')
+            cmd = self._build_package_command(
+                video_file=processed_video_file,
+                audio_file=audio_file,
+                output_file=output_file,
+                start_sec=start_sec,
+                duration_sec=duration_sec
+            )
+
+            # 6. 执行 FFmpeg
+            if not self.run_ffmpeg(cmd, task.task_id):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "TS packaging failed"
+                )
+
+            # 7. 验证输出文件
+            if not self.ensure_file_exists(output_file, min_size=1024):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "TS output file is missing or too small"
+                )
+
+            # 8. 获取实际时长（用于 EXTINF）
+            actual_duration = self.probe_duration(output_file)
+            extinf_duration = actual_duration if actual_duration else duration_sec
+
+            # 9. 上传产物
+            ts_url = self.upload_file(task.task_id, 'ts', output_file)
+            if not ts_url:
+                return TaskResult.fail(
+                    ErrorCode.E_UPLOAD_FAILED,
+                    "Failed to upload TS"
+                )
+
+            return TaskResult.ok({
+                'tsUrl': ts_url,
+                'extinfDurationSec': extinf_duration
+            })
+
+        except Exception as e:
+            logger.error(f"[task:{task.task_id}] Unexpected error: {e}", exc_info=True)
+            return TaskResult.fail(ErrorCode.E_UNKNOWN, str(e))
+
+        finally:
+            self.cleanup_work_dir(work_dir)
+
+    def _build_trim_command(
+        self,
+        video_file: str,
+        output_file: str,
+        trim_head_ms: int,
+        trim_tail_ms: int,
+        output_spec
+    ) -> List[str]:
+        """
+        构建视频精确裁剪命令（重编码方式）
+
+        使用 trim 滤镜进行精确帧级裁剪，而非 -ss/-t 参数的关键帧裁剪。
+
+        Args:
+            video_file: 输入视频路径
+            output_file: 输出视频路径
+            trim_head_ms: 头部裁剪时长（毫秒）
+            trim_tail_ms: 尾部裁剪时长（毫秒）
+            output_spec: 输出规格
+
+        Returns:
+            FFmpeg 命令参数列表
+        """
+        # 获取原视频时长
+        original_duration = self.probe_duration(video_file)
+        if not original_duration:
+            original_duration = 10.0  # 默认值，避免除零
+
+        trim_head_sec = trim_head_ms / 1000.0
+        trim_tail_sec = trim_tail_ms / 1000.0
+
+        # 计算裁剪后的起止时间
+        start_time = trim_head_sec
+        end_time = original_duration - trim_tail_sec
+
+        # 构建 trim 滤镜
+        vf_filter = f"trim=start={start_time}:end={end_time},setpts=PTS-STARTPTS"
+
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            '-i', video_file,
+            '-vf', vf_filter,
+        ]
+
+        # 编码参数
+        cmd.extend(VIDEO_ENCODE_ARGS)
+
+        # 帧率
+        fps = output_spec.fps
+        cmd.extend(['-r', str(fps)])
+
+        # 计算输出视频帧数，动态调整 GOP
+        output_duration_sec = end_time - start_time
+        total_frames = int(output_duration_sec * fps)
+
+        # 动态 GOP：短视频使用较小的 GOP
+        if total_frames <= 1:
+            gop_size = 1
+        elif total_frames < fps:
+            gop_size = total_frames
+        else:
+            gop_size = fps  # 每秒一个关键帧
+
+        cmd.extend(['-g', str(gop_size)])
+        cmd.extend(['-keyint_min', str(min(gop_size, fps // 2 or 1))])
+
+        # 强制第一帧为关键帧
+        cmd.extend(['-force_key_frames', 'expr:eq(n,0)'])
+
+        # 无音频（音频单独处理）
+        cmd.append('-an')
+
+        cmd.append(output_file)
+
+        return cmd
+
+    def _build_package_command(
+        self,
+        video_file: str,
+        audio_file: str,
+        output_file: str,
+        start_sec: float,
+        duration_sec: float
+    ) -> List[str]:
+        """
+        构建 TS 封装命令
+
+        将视频和对应时间区间的音频封装为 TS 分片。
+        视频使用 copy 模式（已经过精确裁剪或无需裁剪）。
+
+        Args:
+            video_file: 视频文件路径（已处理）
+            audio_file: 音频文件路径
+            output_file: 输出文件路径
+            start_sec: 音频开始时间（秒）
+            duration_sec: 音频时长（秒）
+
+        Returns:
+            FFmpeg 命令参数列表
+        """
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            # 视频输入
+            '-i', video_file,
+            # 音频输入（从 start_sec 开始截取 duration_sec）
+            '-ss', str(start_sec),
+            '-t', str(duration_sec),
+            '-i', audio_file,
+            # 映射流
+            '-map', '0:v:0',  # 使用第一个输入的视频流
+            '-map', '1:a:0',  # 使用第二个输入的音频流
+            # 复制编码（视频已处理，无需重编码）
+            '-c:v', 'copy',
+            '-c:a', 'copy',
+            # 关键：时间戳偏移，保证整体连续
+            '-output_ts_offset', str(start_sec),
+            # 复用参数
+            '-muxdelay', '0',
+            '-muxpreload', '0',
+            # 输出格式
+            '-f', 'mpegts',
+            output_file
+        ]
+
+        return cmd

handlers/prepare_audio.py

@@ -0,0 +1,251 @@
+# -*- coding: utf-8 -*-
+"""
+全局音频准备处理器
+
+处理 PREPARE_JOB_AUDIO 任务，生成整个视频的连续音频轨道。
+"""
+
+import os
+import logging
+from typing import List, Dict, Optional
+
+from handlers.base import BaseHandler, AUDIO_ENCODE_ARGS
+from domain.task import Task, TaskType, AudioSpec, AudioProfile
+from domain.result import TaskResult, ErrorCode
+
+logger = logging.getLogger(__name__)
+
+
+class PrepareJobAudioHandler(BaseHandler):
+    """
+    全局音频准备处理器
+
+    职责：
+    - 下载全局 BGM
+    - 下载各片段叠加音效
+    - 构建复杂混音命令
+    - 执行混音
+    - 上传音频产物
+
+    关键约束：
+    - 全局 BGM 连续生成一次，贯穿整个时长
+    - 禁止使用 amix normalize=1
+    - 只对叠加音轨做极短淡入淡出（5-20ms）
+    - 不对 BGM 做边界 fade
+    """
+
+    def get_supported_type(self) -> TaskType:
+        return TaskType.PREPARE_JOB_AUDIO
+
+    def handle(self, task: Task) -> TaskResult:
+        """处理音频准备任务"""
+        work_dir = self.create_work_dir(task.task_id)
+
+        try:
+            # 解析参数
+            total_duration_ms = task.get_total_duration_ms()
+            if total_duration_ms <= 0:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Invalid totalDurationMs"
+                )
+
+            total_duration_sec = total_duration_ms / 1000.0
+            audio_profile = task.get_audio_profile()
+            bgm_url = task.get_bgm_url()
+            segments = task.get_segments()
+
+            # 1. 下载 BGM（如有）
+            bgm_file = None
+            if bgm_url:
+                bgm_file = os.path.join(work_dir, 'bgm.mp3')
+                if not self.download_file(bgm_url, bgm_file):
+                    logger.warning(f"[task:{task.task_id}] Failed to download BGM")
+                    bgm_file = None
+
+            # 2. 下载叠加音效
+            sfx_files = []
+            for i, seg in enumerate(segments):
+                audio_spec_data = seg.get('audioSpecJson')
+                if audio_spec_data:
+                    audio_spec = AudioSpec.from_dict(audio_spec_data)
+                    if audio_spec and audio_spec.audio_url:
+                        sfx_file = os.path.join(work_dir, f'sfx_{i}.mp3')
+                        if self.download_file(audio_spec.audio_url, sfx_file):
+                            sfx_files.append({
+                                'file': sfx_file,
+                                'spec': audio_spec,
+                                'segment': seg
+                            })
+                        else:
+                            logger.warning(f"[task:{task.task_id}] Failed to download SFX {i}")
+
+            # 3. 构建音频混音命令
+            output_file = os.path.join(work_dir, 'audio_full.aac')
+            cmd = self._build_audio_command(
+                bgm_file=bgm_file,
+                sfx_files=sfx_files,
+                output_file=output_file,
+                total_duration_sec=total_duration_sec,
+                audio_profile=audio_profile
+            )
+
+            # 4. 执行 FFmpeg
+            if not self.run_ffmpeg(cmd, task.task_id):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "Audio mixing failed"
+                )
+
+            # 5. 验证输出文件
+            if not self.ensure_file_exists(output_file, min_size=1024):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "Audio output file is missing or too small"
+                )
+
+            # 6. 上传产物
+            audio_url = self.upload_file(task.task_id, 'audio', output_file)
+            if not audio_url:
+                return TaskResult.fail(
+                    ErrorCode.E_UPLOAD_FAILED,
+                    "Failed to upload audio"
+                )
+
+            return TaskResult.ok({
+                'audioUrl': audio_url
+            })
+
+        except Exception as e:
+            logger.error(f"[task:{task.task_id}] Unexpected error: {e}", exc_info=True)
+            return TaskResult.fail(ErrorCode.E_UNKNOWN, str(e))
+
+        finally:
+            self.cleanup_work_dir(work_dir)
+
+    def _build_audio_command(
+        self,
+        bgm_file: Optional[str],
+        sfx_files: List[Dict],
+        output_file: str,
+        total_duration_sec: float,
+        audio_profile: AudioProfile
+    ) -> List[str]:
+        """
+        构建音频混音命令
+
+        Args:
+            bgm_file: BGM 文件路径（可选）
+            sfx_files: 叠加音效列表
+            output_file: 输出文件路径
+            total_duration_sec: 总时长（秒）
+            audio_profile: 音频配置
+
+        Returns:
+            FFmpeg 命令参数列表
+        """
+        sample_rate = audio_profile.sample_rate
+        channels = audio_profile.channels
+
+        # 情况1：无 BGM 也无叠加音效 -> 生成静音
+        if not bgm_file and not sfx_files:
+            return [
+                'ffmpeg', '-y', '-hide_banner',
+                '-f', 'lavfi',
+                '-i', f'anullsrc=r={sample_rate}:cl=stereo',
+                '-t', str(total_duration_sec),
+                '-c:a', 'aac', '-b:a', '128k',
+                output_file
+            ]
+
+        # 情况2：仅 BGM，无叠加音效
+        if not sfx_files:
+            return [
+                'ffmpeg', '-y', '-hide_banner',
+                '-i', bgm_file,
+                '-t', str(total_duration_sec),
+                '-c:a', 'aac', '-b:a', '128k',
+                '-ar', str(sample_rate), '-ac', str(channels),
+                output_file
+            ]
+
+        # 情况3：BGM + 叠加音效 -> 复杂滤镜
+        inputs = []
+        if bgm_file:
+            inputs.extend(['-i', bgm_file])
+        for sfx in sfx_files:
+            inputs.extend(['-i', sfx['file']])
+
+        filter_parts = []
+        input_idx = 0
+
+        # BGM 处理（或生成静音底轨）
+        if bgm_file:
+            filter_parts.append(
+                f"[0:a]atrim=0:{total_duration_sec},asetpts=PTS-STARTPTS,"
+                f"apad=whole_dur={total_duration_sec}[bgm]"
+            )
+            input_idx = 1
+        else:
+            filter_parts.append(
+                f"anullsrc=r={sample_rate}:cl=stereo,"
+                f"atrim=0:{total_duration_sec}[bgm]"
+            )
+            input_idx = 0
+
+        # 叠加音效处理
+        sfx_labels = []
+        for i, sfx in enumerate(sfx_files):
+            idx = input_idx + i
+            spec = sfx['spec']
+            seg = sfx['segment']
+
+            # 计算时间参数
+            start_time_ms = seg.get('startTimeMs', 0)
+            duration_ms = seg.get('durationMs', 5000)
+            delay_ms = start_time_ms + spec.delay_ms
+            delay_sec = delay_ms / 1000.0
+            duration_sec = duration_ms / 1000.0
+
+            # 淡入淡出参数（极短，5-20ms）
+            fade_in_sec = spec.fade_in_ms / 1000.0
+            fade_out_sec = spec.fade_out_ms / 1000.0
+
+            # 音量
+            volume = spec.volume
+
+            label = f"sfx{i}"
+            sfx_labels.append(f"[{label}]")
+
+            # 构建滤镜：延迟 + 淡入淡出 + 音量
+            # 注意：只对叠加音轨做淡入淡出，不对 BGM 做
+            sfx_filter = (
+                f"[{idx}:a]"
+                f"adelay={int(delay_ms)}|{int(delay_ms)},"
+                f"afade=t=in:st={delay_sec}:d={fade_in_sec},"
+                f"afade=t=out:st={delay_sec + duration_sec - fade_out_sec}:d={fade_out_sec},"
+                f"volume={volume}"
+                f"[{label}]"
+            )
+            filter_parts.append(sfx_filter)
+
+        # 混音（关键：normalize=0，禁止归一化）
+        # dropout_transition=0 表示输入结束时不做渐变
+        mix_inputs = "[bgm]" + "".join(sfx_labels)
+        num_inputs = 1 + len(sfx_files)
+        filter_parts.append(
+            f"{mix_inputs}amix=inputs={num_inputs}:duration=first:"
+            f"dropout_transition=0:normalize=0[out]"
+        )
+
+        filter_complex = ';'.join(filter_parts)
+
+        cmd = ['ffmpeg', '-y', '-hide_banner'] + inputs + [
+            '-filter_complex', filter_complex,
+            '-map', '[out]',
+            '-c:a', 'aac', '-b:a', '128k',
+            '-ar', str(sample_rate), '-ac', str(channels),
+            output_file
+        ]
+
+        return cmd

handlers/render_video.py

@@ -0,0 +1,619 @@
+# -*- coding: utf-8 -*-
+"""
+视频片段渲染处理器
+
+处理 RENDER_SEGMENT_VIDEO 任务，将原素材渲染为符合输出规格的视频片段。
+支持转场 overlap 区域的帧冻结生成。
+"""
+
+import os
+import logging
+from typing import List, Optional, Tuple
+from urllib.parse import urlparse, unquote
+
+from handlers.base import BaseHandler
+from domain.task import Task, TaskType, RenderSpec, OutputSpec, Effect, IMAGE_EXTENSIONS
+from domain.result import TaskResult, ErrorCode
+
+logger = logging.getLogger(__name__)
+
+
+def _get_extension_from_url(url: str) -> str:
+    """从 URL 提取文件扩展名"""
+    parsed = urlparse(url)
+    path = unquote(parsed.path)
+    _, ext = os.path.splitext(path)
+    return ext.lower() if ext else ''
+
+
+class RenderSegmentVideoHandler(BaseHandler):
+    """
+    视频片段渲染处理器
+
+    职责：
+    - 下载素材文件
+    - 下载 LUT 文件（如有）
+    - 下载叠加层（如有）
+    - 构建 FFmpeg 渲染命令
+    - 执行渲染（支持帧冻结生成 overlap 区域）
+    - 上传产物
+    """
+
+    def get_supported_type(self) -> TaskType:
+        return TaskType.RENDER_SEGMENT_VIDEO
+
+    def handle(self, task: Task) -> TaskResult:
+        """处理视频渲染任务"""
+        work_dir = self.create_work_dir(task.task_id)
+
+        try:
+            # 解析参数
+            material_url = task.get_material_url()
+            if not material_url:
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    "Missing material URL (boundMaterialUrl or sourceRef)"
+                )
+
+            # 检查 URL 格式：必须是 HTTP 或 HTTPS 协议
+            if not material_url.startswith(('http://', 'https://')):
+                source_ref = task.get_source_ref()
+                bound_url = task.get_bound_material_url()
+                logger.error(
+                    f"[task:{task.task_id}] Invalid material URL format: '{material_url}'. "
+                    f"boundMaterialUrl={bound_url}, sourceRef={source_ref}. "
+                    f"Server should provide boundMaterialUrl with HTTP/HTTPS URL."
+                )
+                return TaskResult.fail(
+                    ErrorCode.E_SPEC_INVALID,
+                    f"Invalid material URL: '{material_url}' is not a valid HTTP/HTTPS URL. "
+                    f"Server must provide boundMaterialUrl."
+                )
+
+            render_spec = task.get_render_spec()
+            output_spec = task.get_output_spec()
+            duration_ms = task.get_duration_ms()
+
+            # 1. 检测素材类型并确定输入文件扩展名
+            is_image = task.is_image_material()
+            if is_image:
+                # 图片素材：根据 URL 确定扩展名
+                ext = _get_extension_from_url(material_url)
+                if not ext or ext not in IMAGE_EXTENSIONS:
+                    ext = '.jpg'  # 默认扩展名
+                input_file = os.path.join(work_dir, f'input{ext}')
+            else:
+                input_file = os.path.join(work_dir, 'input.mp4')
+
+            # 2. 下载素材
+            if not self.download_file(material_url, input_file):
+                return TaskResult.fail(
+                    ErrorCode.E_INPUT_UNAVAILABLE,
+                    f"Failed to download material: {material_url}"
+                )
+
+            # 3. 图片素材转换为视频
+            if is_image:
+                video_input_file = os.path.join(work_dir, 'input_video.mp4')
+                if not self._convert_image_to_video(
+                    image_file=input_file,
+                    output_file=video_input_file,
+                    duration_ms=duration_ms,
+                    output_spec=output_spec,
+                    render_spec=render_spec,
+                    task_id=task.task_id
+                ):
+                    return TaskResult.fail(
+                        ErrorCode.E_FFMPEG_FAILED,
+                        "Failed to convert image to video"
+                    )
+                # 使用转换后的视频作为输入
+                input_file = video_input_file
+                logger.info(f"[task:{task.task_id}] Image converted to video successfully")
+
+            # 4. 下载 LUT（如有）
+            lut_file = None
+            if render_spec.lut_url:
+                lut_file = os.path.join(work_dir, 'lut.cube')
+                if not self.download_file(render_spec.lut_url, lut_file):
+                    logger.warning(f"[task:{task.task_id}] Failed to download LUT, continuing without it")
+                    lut_file = None
+
+            # 5. 下载叠加层（如有）
+            overlay_file = None
+            if render_spec.overlay_url:
+                # 根据 URL 后缀确定文件扩展名
+                url_lower = render_spec.overlay_url.lower()
+                if url_lower.endswith('.jpg') or url_lower.endswith('.jpeg'):
+                    ext = '.jpg'
+                elif url_lower.endswith('.mov'):
+                    ext = '.mov'
+                else:
+                    ext = '.png'  # 默认
+                overlay_file = os.path.join(work_dir, f'overlay{ext}')
+                if not self.download_file(render_spec.overlay_url, overlay_file):
+                    logger.warning(f"[task:{task.task_id}] Failed to download overlay, continuing without it")
+                    overlay_file = None
+
+            # 6. 计算 overlap 时长（用于转场帧冻结）
+            # 头部 overlap: 来自前一片段的出场转场
+            overlap_head_ms = task.get_overlap_head_ms()
+            # 尾部 overlap: 当前片段的出场转场
+            overlap_tail_ms = task.get_overlap_tail_ms_v2()
+
+            # 7. 构建 FFmpeg 命令
+            output_file = os.path.join(work_dir, 'output.mp4')
+            cmd = self._build_command(
+                input_file=input_file,
+                output_file=output_file,
+                render_spec=render_spec,
+                output_spec=output_spec,
+                duration_ms=duration_ms,
+                lut_file=lut_file,
+                overlay_file=overlay_file,
+                overlap_head_ms=overlap_head_ms,
+                overlap_tail_ms=overlap_tail_ms
+            )
+
+            # 8. 执行 FFmpeg
+            if not self.run_ffmpeg(cmd, task.task_id):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "FFmpeg rendering failed"
+                )
+
+            # 9. 验证输出文件
+            if not self.ensure_file_exists(output_file, min_size=4096):
+                return TaskResult.fail(
+                    ErrorCode.E_FFMPEG_FAILED,
+                    "Output file is missing or too small"
+                )
+
+            # 10. 获取实际时长
+            actual_duration = self.probe_duration(output_file)
+            actual_duration_ms = int(actual_duration * 1000) if actual_duration else duration_ms
+
+            # 11. 上传产物
+            video_url = self.upload_file(task.task_id, 'video', output_file)
+            if not video_url:
+                return TaskResult.fail(
+                    ErrorCode.E_UPLOAD_FAILED,
+                    "Failed to upload video"
+                )
+
+            # 12. 构建结果（包含 overlap 信息）
+            result_data = {
+                'videoUrl': video_url,
+                'actualDurationMs': actual_duration_ms,
+                'overlapHeadMs': overlap_head_ms,
+                'overlapTailMs': overlap_tail_ms
+            }
+
+            return TaskResult.ok(result_data)
+
+        except Exception as e:
+            logger.error(f"[task:{task.task_id}] Unexpected error: {e}", exc_info=True)
+            return TaskResult.fail(ErrorCode.E_UNKNOWN, str(e))
+
+        finally:
+            self.cleanup_work_dir(work_dir)
+
+    def _convert_image_to_video(
+        self,
+        image_file: str,
+        output_file: str,
+        duration_ms: int,
+        output_spec: OutputSpec,
+        render_spec: RenderSpec,
+        task_id: str
+    ) -> bool:
+        """
+        将图片转换为视频
+
+        使用 FFmpeg 将静态图片转换为指定时长的视频，
+        同时应用缩放填充和变速处理。
+
+        Args:
+            image_file: 输入图片文件路径
+            output_file: 输出视频文件路径
+            duration_ms: 目标时长（毫秒）
+            output_spec: 输出规格
+            render_spec: 渲染规格
+            task_id: 任务 ID（用于日志）
+
+        Returns:
+            是否成功
+        """
+        width = output_spec.width
+        height = output_spec.height
+        fps = output_spec.fps
+
+        # 计算实际时长（考虑变速）
+        speed = float(render_spec.speed) if render_spec.speed else 1.0
+        if speed <= 0:
+            speed = 1.0
+        # 变速后的实际播放时长
+        actual_duration_sec = (duration_ms / 1000.0) / speed
+
+        # 构建 FFmpeg 命令
+        cmd = [
+            'ffmpeg', '-y', '-hide_banner',
+            '-loop', '1',  # 循环输入图片
+            '-i', image_file,
+            '-t', str(actual_duration_sec),  # 输出时长
+        ]
+
+        # 构建滤镜：缩放填充到目标尺寸
+        filters = []
+
+        # 裁切处理（与视频相同逻辑）
+        if render_spec.crop_enable and render_spec.face_pos:
+            try:
+                fx, fy = map(float, render_spec.face_pos.split(','))
+                target_ratio = width / height
+                filters.append(
+                    f"crop='min(iw,ih*{target_ratio})':'min(ih,iw/{target_ratio})':"
+                    f"'(iw-min(iw,ih*{target_ratio}))*{fx}':"
+                    f"'(ih-min(ih,iw/{target_ratio}))*{fy}'"
+                )
+            except (ValueError, ZeroDivisionError):
+                logger.warning(f"[task:{task_id}] Invalid face position: {render_spec.face_pos}")
+        elif render_spec.zoom_cut:
+            target_ratio = width / height
+            filters.append(
+                f"crop='min(iw,ih*{target_ratio})':'min(ih,iw/{target_ratio})'"
+            )
+
+        # 缩放填充
+        filters.append(
+            f"scale={width}:{height}:force_original_aspect_ratio=decrease,"
+            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black"
+        )
+
+        # 格式转换（确保兼容性）
+        filters.append("format=yuv420p")
+
+        cmd.extend(['-vf', ','.join(filters)])
+
+        # 编码参数
+        cmd.extend([
+            '-c:v', 'libx264',
+            '-preset', 'fast',
+            '-crf', '18',
+            '-r', str(fps),
+            '-an',  # 无音频
+            output_file
+        ])
+
+        logger.info(f"[task:{task_id}] Converting image to video: {actual_duration_sec:.2f}s at {fps}fps")
+        return self.run_ffmpeg(cmd, task_id)
+
+    def _build_command(
+        self,
+        input_file: str,
+        output_file: str,
+        render_spec: RenderSpec,
+        output_spec: OutputSpec,
+        duration_ms: int,
+        lut_file: Optional[str] = None,
+        overlay_file: Optional[str] = None,
+        overlap_head_ms: int = 0,
+        overlap_tail_ms: int = 0
+    ) -> List[str]:
+        """
+        构建 FFmpeg 渲染命令
+
+        Args:
+            input_file: 输入文件路径
+            output_file: 输出文件路径
+            render_spec: 渲染规格
+            output_spec: 输出规格
+            duration_ms: 目标时长（毫秒）
+            lut_file: LUT 文件路径（可选）
+            overlay_file: 叠加层文件路径（可选）
+            overlap_head_ms: 头部 overlap 时长（毫秒）
+            overlap_tail_ms: 尾部 overlap 时长（毫秒）
+
+        Returns:
+            FFmpeg 命令参数列表
+        """
+        cmd = ['ffmpeg', '-y', '-hide_banner']
+
+        # 硬件加速解码参数（在输入文件之前）
+        hwaccel_args = self.get_hwaccel_decode_args()
+        if hwaccel_args:
+            cmd.extend(hwaccel_args)
+
+        # 输入文件
+        cmd.extend(['-i', input_file])
+
+        # 叠加层输入
+        if overlay_file:
+            cmd.extend(['-i', overlay_file])
+
+        # 构建视频滤镜链
+        filters = self._build_video_filters(
+            render_spec=render_spec,
+            output_spec=output_spec,
+            lut_file=lut_file,
+            overlay_file=overlay_file,
+            overlap_head_ms=overlap_head_ms,
+            overlap_tail_ms=overlap_tail_ms
+        )
+
+        # 应用滤镜
+        # 检测是否为 filter_complex 格式（包含分号或方括号标签）
+        is_filter_complex = ';' in filters or (filters.startswith('[') and ']' in filters)
+        if is_filter_complex or overlay_file:
+            # 使用 filter_complex 处理
+            cmd.extend(['-filter_complex', filters])
+        elif filters:
+            cmd.extend(['-vf', filters])
+
+        # 编码参数（根据硬件加速配置动态获取）
+        cmd.extend(self.get_video_encode_args())
+
+        # 帧率
+        fps = output_spec.fps
+        cmd.extend(['-r', str(fps)])
+
+        # GOP 大小（关键帧间隔）
+        gop_size = fps * 2  # 2秒一个关键帧
+        cmd.extend(['-g', str(gop_size)])
+        cmd.extend(['-keyint_min', str(gop_size)])
+
+        # 时长（包含 overlap 区域）
+        total_duration_ms = duration_ms + overlap_head_ms + overlap_tail_ms
+        duration_sec = total_duration_ms / 1000.0
+        cmd.extend(['-t', str(duration_sec)])
+
+        # 无音频（视频片段不包含音频）
+        cmd.append('-an')
+
+        # 输出文件
+        cmd.append(output_file)
+
+        return cmd
+
+    def _build_video_filters(
+        self,
+        render_spec: RenderSpec,
+        output_spec: OutputSpec,
+        lut_file: Optional[str] = None,
+        overlay_file: Optional[str] = None,
+        overlap_head_ms: int = 0,
+        overlap_tail_ms: int = 0
+    ) -> str:
+        """
+        构建视频滤镜链
+
+        Args:
+            render_spec: 渲染规格
+            output_spec: 输出规格
+            lut_file: LUT 文件路径
+            overlay_file: 叠加层文件路径（支持图片 png/jpg 和视频 mov）
+            overlap_head_ms: 头部 overlap 时长（毫秒）
+            overlap_tail_ms: 尾部 overlap 时长（毫秒）
+
+        Returns:
+            滤镜字符串
+        """
+        filters = []
+        width = output_spec.width
+        height = output_spec.height
+        fps = output_spec.fps
+
+        # 判断 overlay 类型
+        has_overlay = overlay_file is not None
+        is_video_overlay = has_overlay and overlay_file.lower().endswith('.mov')
+
+        # 解析 effects
+        effects = render_spec.get_effects()
+        has_camera_shot = any(e.effect_type == 'cameraShot' for e in effects)
+
+        # 硬件加速时需要先 hwdownload（将 GPU 表面下载到系统内存）
+        hwaccel_prefix = self.get_hwaccel_filter_prefix()
+        if hwaccel_prefix:
+            # 去掉末尾的逗号，作为第一个滤镜
+            filters.append(hwaccel_prefix.rstrip(','))
+
+        # 1. 变速处理
+        speed = float(render_spec.speed) if render_spec.speed else 1.0
+        if speed != 1.0 and speed > 0:
+            # setpts 公式：PTS / speed
+            pts_factor = 1.0 / speed
+            filters.append(f"setpts={pts_factor}*PTS")
+
+        # 2. LUT 调色
+        if lut_file:
+            # 路径中的反斜杠需要转换，冒号需要转义（FFmpeg filter语法中冒号是特殊字符）
+            lut_path = lut_file.replace('\\', '/').replace(':', r'\:')
+            filters.append(f"lut3d='{lut_path}'")
+
+        # 3. 裁切处理
+        if render_spec.crop_enable and render_spec.face_pos:
+            # 根据人脸位置进行智能裁切
+            try:
+                fx, fy = map(float, render_spec.face_pos.split(','))
+                # 计算裁切区域（保持输出比例）
+                target_ratio = width / height
+                # 假设裁切到目标比例
+                filters.append(
+                    f"crop='min(iw,ih*{target_ratio})':'min(ih,iw/{target_ratio})':"
+                    f"'(iw-min(iw,ih*{target_ratio}))*{fx}':"
+                    f"'(ih-min(ih,iw/{target_ratio}))*{fy}'"
+                )
+            except (ValueError, ZeroDivisionError):
+                logger.warning(f"Invalid face position: {render_spec.face_pos}")
+        elif render_spec.zoom_cut:
+            # 中心缩放裁切
+            target_ratio = width / height
+            filters.append(
+                f"crop='min(iw,ih*{target_ratio})':'min(ih,iw/{target_ratio})'"
+            )
+
+        # 4. 缩放和填充
+        scale_filter = (
+            f"scale={width}:{height}:force_original_aspect_ratio=decrease,"
+            f"pad={width}:{height}:(ow-iw)/2:(oh-ih)/2:black"
+        )
+        filters.append(scale_filter)
+
+        # 5. 特效处理（cameraShot 需要特殊处理）
+        if has_camera_shot:
+            # cameraShot 需要使用 filter_complex 格式
+            return self._build_filter_complex_with_effects(
+                base_filters=filters,
+                effects=effects,
+                fps=fps,
+                has_overlay=has_overlay,
+                is_video_overlay=is_video_overlay,
+                overlap_head_ms=overlap_head_ms,
+                overlap_tail_ms=overlap_tail_ms,
+                use_hwdownload=bool(hwaccel_prefix)
+            )
+
+        # 6. 帧冻结（tpad）- 用于转场 overlap 区域
+        # 注意：tpad 必须在缩放之后应用
+        tpad_parts = []
+        if overlap_head_ms > 0:
+            # 头部冻结：将第一帧冻结指定时长
+            head_duration_sec = overlap_head_ms / 1000.0
+            tpad_parts.append(f"start_mode=clone:start_duration={head_duration_sec}")
+        if overlap_tail_ms > 0:
+            # 尾部冻结：将最后一帧冻结指定时长
+            tail_duration_sec = overlap_tail_ms / 1000.0
+            tpad_parts.append(f"stop_mode=clone:stop_duration={tail_duration_sec}")
+
+        if tpad_parts:
+            filters.append(f"tpad={':'.join(tpad_parts)}")
+
+        # 7. 构建最终滤镜
+        if has_overlay:
+            # 使用 filter_complex 格式
+            base_filters = ','.join(filters) if filters else 'copy'
+            # 视频 overlay 使用 eof_action=pass（结束后消失），图片 overlay 使用默认行为（保持显示）
+            overlay_params = 'eof_action=pass' if is_video_overlay else ''
+            overlay_filter = f"overlay=0:0:{overlay_params}" if overlay_params else 'overlay=0:0'
+            # 视频 overlay 需要在末尾统一颜色范围，避免 overlay 结束后 range 从 tv 变为 pc
+            range_fix = ',format=yuv420p,setrange=tv' if is_video_overlay else ''
+            return f"[0:v]{base_filters}[base];[base][1:v]{overlay_filter}{range_fix}"
+        else:
+            return ','.join(filters) if filters else ''
+
+    def _build_filter_complex_with_effects(
+        self,
+        base_filters: List[str],
+        effects: List[Effect],
+        fps: int,
+        has_overlay: bool = False,
+        is_video_overlay: bool = False,
+        overlap_head_ms: int = 0,
+        overlap_tail_ms: int = 0,
+        use_hwdownload: bool = False
+    ) -> str:
+        """
+        构建包含特效的 filter_complex 滤镜图
+
+        cameraShot 效果需要使用 split/freezeframes/concat 滤镜组合。
+
+        Args:
+            base_filters: 基础滤镜列表
+            effects: 特效列表
+            fps: 帧率
+            has_overlay: 是否有叠加层
+            is_video_overlay: 叠加层是否为视频格式（如 .mov）
+            overlap_head_ms: 头部 overlap 时长
+            overlap_tail_ms: 尾部 overlap 时长
+            use_hwdownload: 是否使用了硬件加速解码（已在 base_filters 中包含 hwdownload）
+
+        Returns:
+            filter_complex 格式的滤镜字符串
+        """
+        filter_parts = []
+
+        # 基础滤镜链
+        base_chain = ','.join(base_filters) if base_filters else 'copy'
+
+        # 当前输出标签
+        current_output = '[v_base]'
+        filter_parts.append(f"[0:v]{base_chain}{current_output}")
+
+        # 处理每个特效
+        effect_idx = 0
+        for effect in effects:
+            if effect.effect_type == 'cameraShot':
+                start_sec, duration_sec = effect.get_camera_shot_params()
+                if start_sec <= 0 or duration_sec <= 0:
+                    continue
+
+                # cameraShot 实现（定格效果）：
+                # 1. fps + split 分割
+                # 2. 第一路：trim(0, start) + tpad冻结duration秒
+                # 3. 第二路：trim(start, end)
+                # 4. concat 拼接
+
+                split_out_a = f'[eff{effect_idx}_a]'
+                split_out_b = f'[eff{effect_idx}_b]'
+                frozen_out = f'[eff{effect_idx}_frozen]'
+                rest_out = f'[eff{effect_idx}_rest]'
+                effect_output = f'[v_eff{effect_idx}]'
+
+                # fps + split
+                filter_parts.append(
+                    f"{current_output}fps=fps={fps},split{split_out_a}{split_out_b}"
+                )
+
+                # 第一路：trim(0, start) + tpad冻结
+                # tpad=stop_mode=clone 将最后一帧冻结指定时长
+                filter_parts.append(
+                    f"{split_out_a}trim=start=0:end={start_sec},setpts=PTS-STARTPTS,"
+                    f"tpad=stop_mode=clone:stop_duration={duration_sec}{frozen_out}"
+                )
+
+                # 第二路：trim 从 start 开始
+                filter_parts.append(
+                    f"{split_out_b}trim=start={start_sec},setpts=PTS-STARTPTS{rest_out}"
+                )
+
+                # concat 拼接
+                filter_parts.append(
+                    f"{frozen_out}{rest_out}concat=n=2:v=1:a=0{effect_output}"
+                )
+
+                current_output = effect_output
+                effect_idx += 1
+
+        # 帧冻结（tpad）- 用于转场 overlap 区域
+        tpad_parts = []
+        if overlap_head_ms > 0:
+            head_duration_sec = overlap_head_ms / 1000.0
+            tpad_parts.append(f"start_mode=clone:start_duration={head_duration_sec}")
+        if overlap_tail_ms > 0:
+            tail_duration_sec = overlap_tail_ms / 1000.0
+            tpad_parts.append(f"stop_mode=clone:stop_duration={tail_duration_sec}")
+
+        if tpad_parts:
+            tpad_output = '[v_tpad]'
+            filter_parts.append(f"{current_output}tpad={':'.join(tpad_parts)}{tpad_output}")
+            current_output = tpad_output
+
+        # 最终输出
+        if has_overlay:
+            # 叠加层处理
+            # 视频 overlay 使用 eof_action=pass（结束后消失），图片 overlay 使用默认行为（保持显示）
+            overlay_params = 'eof_action=pass' if is_video_overlay else ''
+            overlay_filter = f"overlay=0:0:{overlay_params}" if overlay_params else 'overlay=0:0'
+            # 视频 overlay 需要在末尾统一颜色范围，避免 overlay 结束后 range 从 tv 变为 pc
+            range_fix = ',format=yuv420p,setrange=tv' if is_video_overlay else ''
+            filter_parts.append(f"{current_output}[1:v]{overlay_filter}{range_fix}")
+        else:
+            # 移除最后一个标签，直接输出
+            # 将最后一个滤镜的输出标签替换为空（直接输出）
+            if filter_parts:
+                last_filter = filter_parts[-1]
+                # 移除末尾的输出标签
+                if last_filter.endswith(current_output):
+                    filter_parts[-1] = last_filter[:-len(current_output)]
+
+        return ';'.join(filter_parts)

index.py

@@ -1,52 +1,182 @@
-from time import sleep
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+RenderWorker v2 入口
+
+支持 v2 API 协议的渲染 Worker，处理以下任务类型：
+- RENDER_SEGMENT_VIDEO: 渲染视频片段
+- PREPARE_JOB_AUDIO: 生成全局音频
+- PACKAGE_SEGMENT_TS: 封装 TS 分片
+- FINALIZE_MP4: 产出最终 MP4
+
+使用方法:
+    python index.py
+
+环境变量:
+    API_ENDPOINT_V2: v2 API 端点（或使用 API_ENDPOINT）
+    ACCESS_KEY: Worker 认证密钥
+    WORKER_ID: Worker ID（默认 100001）
+    MAX_CONCURRENCY: 最大并发数（默认 4）
+    HEARTBEAT_INTERVAL: 心跳间隔秒数（默认 5）
+    TEMP_DIR: 临时文件目录
+"""
+
 import sys
-
+import time
-import config
+import signal
-import biz.task
-from telemetry import init_opentelemetry
-from template import load_local_template, download_template, TEMPLATES
-from util import api
-
-import os
-import glob
-
-load_local_template()
-
-# Check for redownload parameter
-if 'redownload' in sys.argv:
-    print("Redownloading all templates...")
-    for template_name in TEMPLATES.keys():
-        print(f"Redownloading template: {template_name}")
-        download_template(template_name)
-    print("All templates redownloaded successfully!")
-    sys.exit(0)
 import logging
 
-LOGGER = logging.getLogger(__name__)
+from dotenv import load_dotenv
-init_opentelemetry()
 
-while True:
+from domain.config import WorkerConfig
-    # print(get_sys_info())
+from services.api_client import APIClientV2
-    print("waiting for task...")
+from services.task_executor import TaskExecutor
-    try:
+from constant import SOFTWARE_VERSION
-        task_list = api.sync_center()
+
-    except Exception as e:
+# 日志配置
-        LOGGER.error("sync_center error", exc_info=e)
+logging.basicConfig(
-        sleep(5)
+    level=logging.INFO,
-        continue
+    format='[%(asctime)s] [%(levelname)s] [%(name)s] %(message)s',
-    if len(task_list) == 0:
+    datefmt='%Y-%m-%d %H:%M:%S'
-        # 删除当前文件夹下所有以.mp4、.ts结尾的文件
+)
-        for file_globs in ['*.mp4', '*.ts', 'tmp_concat*.txt']:
+logger = logging.getLogger('worker')
-            for file_path in glob.glob(file_globs):
+
-                try:
+
-                    os.remove(file_path)
+class WorkerV2:
-                    print(f"Deleted file: {file_path}")
+    """
-                except Exception as e:
+    v2 渲染 Worker 主类
-                    LOGGER.error(f"Error deleting file {file_path}", exc_info=e)
+
-        sleep(5)
+    负责：
-    for task in task_list:
+    - 配置加载
-        print("start task:", task)
+    - API 客户端初始化
+    - 任务执行器管理
+    - 主循环运行
+    - 优雅退出处理
+    """
+
+    def __init__(self):
+        """初始化 Worker"""
+        # 加载配置
         try:
-            biz.task.start_task(task)
+            self.config = WorkerConfig.from_env()
-        except Exception as e:
+        except ValueError as e:
-            LOGGER.error("task_start error", exc_info=e)
+            logger.error(f"Configuration error: {e}")
+            sys.exit(1)
+
+        # 初始化 API 客户端
+        self.api_client = APIClientV2(self.config)
+
+        # 初始化任务执行器
+        self.task_executor = TaskExecutor(self.config, self.api_client)
+
+        # 运行状态
+        self.running = True
+
+        # 确保临时目录存在
+        self.config.ensure_temp_dir()
+
+        # 注册信号处理器
+        self._setup_signal_handlers()
+
+    def _setup_signal_handlers(self):
+        """设置信号处理器"""
+        # Windows 不支持 SIGTERM
+        signal.signal(signal.SIGINT, self._signal_handler)
+        if hasattr(signal, 'SIGTERM'):
+            signal.signal(signal.SIGTERM, self._signal_handler)
+
+    def _signal_handler(self, signum, frame):
+        """
+        信号处理，优雅退出
+
+        Args:
+            signum: 信号编号
+            frame: 当前栈帧
+        """
+        signal_name = signal.Signals(signum).name
+        logger.info(f"Received signal {signal_name}, initiating shutdown...")
+        self.running = False
+
+    def run(self):
+        """主循环"""
+        logger.info("=" * 60)
+        logger.info("RenderWorker v2 Starting")
+        logger.info("=" * 60)
+        logger.info(f"Worker ID: {self.config.worker_id}")
+        logger.info(f"API Endpoint: {self.config.api_endpoint}")
+        logger.info(f"Max Concurrency: {self.config.max_concurrency}")
+        logger.info(f"Heartbeat Interval: {self.config.heartbeat_interval}s")
+        logger.info(f"Capabilities: {', '.join(self.config.capabilities)}")
+        logger.info(f"Temp Directory: {self.config.temp_dir}")
+        logger.info("=" * 60)
+
+        consecutive_errors = 0
+        max_consecutive_errors = 10
+
+        while self.running:
+            try:
+                # 心跳同步并拉取任务
+                current_task_ids = self.task_executor.get_current_task_ids()
+                tasks = self.api_client.sync(current_task_ids)
+
+                # 提交新任务
+                for task in tasks:
+                    if self.task_executor.submit_task(task):
+                        logger.info(f"Submitted task: {task.task_id} ({task.task_type.value})")
+
+                # 重置错误计数
+                consecutive_errors = 0
+
+                # 等待下次心跳
+                time.sleep(self.config.heartbeat_interval)
+
+            except KeyboardInterrupt:
+                logger.info("Keyboard interrupt received")
+                self.running = False
+            except Exception as e:
+                consecutive_errors += 1
+                logger.error(f"Worker loop error ({consecutive_errors}/{max_consecutive_errors}): {e}", exc_info=True)
+
+                # 连续错误过多，增加等待时间
+                if consecutive_errors >= max_consecutive_errors:
+                    logger.error("Too many consecutive errors, waiting 30 seconds...")
+                    time.sleep(30)
+                    consecutive_errors = 0
+                else:
+                    time.sleep(5)
+
+        # 优雅关闭
+        self._shutdown()
+
+    def _shutdown(self):
+        """优雅关闭"""
+        logger.info("Shutting down...")
+
+        # 等待当前任务完成
+        current_count = self.task_executor.get_current_task_count()
+        if current_count > 0:
+            logger.info(f"Waiting for {current_count} running task(s) to complete...")
+
+        # 关闭执行器
+        self.task_executor.shutdown(wait=True)
+
+        # 关闭 API 客户端
+        self.api_client.close()
+
+        logger.info("Worker stopped")
+
+
+def main():
+    """主函数"""
+    # 加载 .env 文件（如果存在）
+    load_dotenv()
+
+    logger.info(f"RenderWorker v{SOFTWARE_VERSION}")
+
+    # 创建并运行 Worker
+    worker = WorkerV2()
+    worker.run()
+
+
+if __name__ == '__main__':
+    main()

services/__init__.py

@@ -0,0 +1,18 @@
+# -*- coding: utf-8 -*-
+"""
+服务层
+
+包含 API 客户端、任务执行器、租约服务、存储服务等组件。
+"""
+
+from services.api_client import APIClientV2
+from services.lease_service import LeaseService
+from services.task_executor import TaskExecutor
+from services import storage
+
+__all__ = [
+    'APIClientV2',
+    'LeaseService',
+    'TaskExecutor',
+    'storage',
+]

services/api_client.py

@@ -0,0 +1,417 @@
+# -*- coding: utf-8 -*-
+"""
+v2 API 客户端
+
+实现与渲染服务端 v2 接口的通信。
+"""
+
+import logging
+import subprocess
+import time
+import requests
+from typing import Dict, List, Optional, Any
+
+from domain.task import Task
+from domain.config import WorkerConfig
+from util.system import get_hw_accel_info_str
+
+logger = logging.getLogger(__name__)
+
+
+class APIClientV2:
+    """
+    v2 API 客户端
+
+    负责与渲染服务端的所有 HTTP 通信。
+    """
+
+    SYSTEM_INFO_TTL_SECONDS = 30
+
+    def __init__(self, config: WorkerConfig):
+        """
+        初始化 API 客户端
+
+        Args:
+            config: Worker 配置
+        """
+        self.config = config
+        self.base_url = config.api_endpoint.rstrip('/')
+        self.access_key = config.access_key
+        self.worker_id = config.worker_id
+        self.session = requests.Session()
+
+        self._ffmpeg_version: Optional[str] = None
+        self._codec_info: Optional[str] = None
+        self._hw_accel_info: Optional[str] = None
+        self._gpu_info: Optional[str] = None
+        self._gpu_info_checked = False
+        self._static_system_info: Optional[Dict[str, Any]] = None
+        self._system_info_cache: Optional[Dict[str, Any]] = None
+        self._system_info_cache_ts = 0.0
+
+        # 设置默认请求头
+        self.session.headers.update({
+            'Content-Type': 'application/json',
+            'Accept': 'application/json'
+        })
+
+    def sync(self, current_task_ids: List[str]) -> List[Task]:
+        """
+        心跳同步并拉取任务
+
+        Args:
+            current_task_ids: 当前正在执行的任务 ID 列表
+
+        Returns:
+            List[Task]: 新分配的任务列表
+        """
+        url = f"{self.base_url}/render/v2/worker/sync"
+
+        # 将 task_id 转换为整数（服务端期望 []int64）
+        task_ids_int = [int(tid) for tid in current_task_ids if tid.isdigit()]
+
+        payload = {
+            'accessKey': self.access_key,
+            'workerId': self.worker_id,
+            'capabilities': self.config.capabilities,
+            'maxConcurrency': self.config.max_concurrency,
+            'currentTaskCount': len(current_task_ids),
+            'currentTaskIds': task_ids_int,
+            'ffmpegVersion': self._get_ffmpeg_version(),
+            'codecInfo': self._get_codec_info(),
+            'systemInfo': self._get_system_info()
+        }
+
+        try:
+            resp = self.session.post(url, json=payload, timeout=10)
+            resp.raise_for_status()
+            data = resp.json()
+
+            if data.get('code') != 200:
+                logger.warning(f"Sync failed: {data.get('message')}")
+                return []
+
+            # 解析任务列表
+            tasks = []
+            for task_data in data.get('data', {}).get('tasks') or []:
+                try:
+                    task = Task.from_dict(task_data)
+                    tasks.append(task)
+                except Exception as e:
+                    logger.error(f"Failed to parse task: {e}")
+
+            if tasks:
+                logger.info(f"Received {len(tasks)} new tasks")
+
+            return tasks
+
+        except requests.exceptions.Timeout:
+            logger.warning("Sync timeout")
+            return []
+        except requests.exceptions.RequestException as e:
+            logger.error(f"Sync request error: {e}")
+            return []
+        except Exception as e:
+            logger.error(f"Sync error: {e}")
+            return []
+
+    def report_start(self, task_id: str) -> bool:
+        """
+        报告任务开始
+
+        Args:
+            task_id: 任务 ID
+
+        Returns:
+            bool: 是否成功
+        """
+        url = f"{self.base_url}/render/v2/task/{task_id}/start"
+
+        try:
+            resp = self.session.post(
+                url,
+                json={'workerId': self.worker_id},
+                timeout=10
+            )
+            if resp.status_code == 200:
+                logger.debug(f"[task:{task_id}] Start reported")
+                return True
+            else:
+                logger.warning(f"[task:{task_id}] Report start failed: {resp.status_code}")
+                return False
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Report start error: {e}")
+            return False
+
+    def report_success(self, task_id: str, result: Dict[str, Any]) -> bool:
+        """
+        报告任务成功
+
+        Args:
+            task_id: 任务 ID
+            result: 任务结果数据
+
+        Returns:
+            bool: 是否成功
+        """
+        url = f"{self.base_url}/render/v2/task/{task_id}/success"
+
+        try:
+            resp = self.session.post(
+                url,
+                json={
+                    'workerId': self.worker_id,
+                    'result': result
+                },
+                timeout=10
+            )
+            if resp.status_code == 200:
+                logger.debug(f"[task:{task_id}] Success reported")
+                return True
+            else:
+                logger.warning(f"[task:{task_id}] Report success failed: {resp.status_code}")
+                return False
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Report success error: {e}")
+            return False
+
+    def report_fail(self, task_id: str, error_code: str, error_message: str) -> bool:
+        """
+        报告任务失败
+
+        Args:
+            task_id: 任务 ID
+            error_code: 错误码
+            error_message: 错误信息
+
+        Returns:
+            bool: 是否成功
+        """
+        url = f"{self.base_url}/render/v2/task/{task_id}/fail"
+
+        try:
+            resp = self.session.post(
+                url,
+                json={
+                    'workerId': self.worker_id,
+                    'errorCode': error_code,
+                    'errorMessage': error_message[:1000]  # 限制长度
+                },
+                timeout=10
+            )
+            if resp.status_code == 200:
+                logger.debug(f"[task:{task_id}] Failure reported")
+                return True
+            else:
+                logger.warning(f"[task:{task_id}] Report fail failed: {resp.status_code}")
+                return False
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Report fail error: {e}")
+            return False
+
+    def get_upload_url(self, task_id: str, file_type: str, file_name: str = None) -> Optional[Dict[str, str]]:
+        """
+        获取上传 URL
+
+        Args:
+            task_id: 任务 ID
+            file_type: 文件类型（video/audio/ts/mp4）
+            file_name: 文件名（可选）
+
+        Returns:
+            Dict 包含 uploadUrl 和 accessUrl，失败返回 None
+        """
+        url = f"{self.base_url}/render/v2/task/{task_id}/uploadUrl"
+
+        payload = {'fileType': file_type}
+        if file_name:
+            payload['fileName'] = file_name
+
+        try:
+            resp = self.session.post(url, json=payload, timeout=10)
+            if resp.status_code == 200:
+                data = resp.json()
+                if data.get('code') == 200:
+                    return data.get('data')
+            logger.warning(f"[task:{task_id}] Get upload URL failed: {resp.status_code}")
+            return None
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Get upload URL error: {e}")
+            return None
+
+    def extend_lease(self, task_id: str, extension: int = None) -> bool:
+        """
+        延长租约
+
+        Args:
+            task_id: 任务 ID
+            extension: 延长秒数（默认使用配置值）
+
+        Returns:
+            bool: 是否成功
+        """
+        if extension is None:
+            extension = self.config.lease_extension_duration
+
+        url = f"{self.base_url}/render/v2/task/{task_id}/extend-lease"
+
+        try:
+            resp = self.session.post(
+                url,
+                params={
+                    'workerId': self.worker_id,
+                    'extension': extension
+                },
+                timeout=10
+            )
+            if resp.status_code == 200:
+                logger.debug(f"[task:{task_id}] Lease extended by {extension}s")
+                return True
+            else:
+                logger.warning(f"[task:{task_id}] Extend lease failed: {resp.status_code}")
+                return False
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Extend lease error: {e}")
+            return False
+
+    def get_task_info(self, task_id: str) -> Optional[Dict]:
+        """
+        获取任务详情
+
+        Args:
+            task_id: 任务 ID
+
+        Returns:
+            任务详情字典，失败返回 None
+        """
+        url = f"{self.base_url}/render/v2/task/{task_id}"
+
+        try:
+            resp = self.session.get(url, timeout=10)
+            if resp.status_code == 200:
+                data = resp.json()
+                if data.get('code') == 200:
+                    return data.get('data')
+            return None
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Get task info error: {e}")
+            return None
+
+    def _get_ffmpeg_version(self) -> str:
+        """获取 FFmpeg 版本"""
+        if self._ffmpeg_version is not None:
+            return self._ffmpeg_version
+        try:
+            result = subprocess.run(
+                ['ffmpeg', '-version'],
+                capture_output=True,
+                text=True,
+                timeout=5
+            )
+            first_line = result.stdout.split('\n')[0]
+            if 'version' in first_line:
+                parts = first_line.split()
+                for i, part in enumerate(parts):
+                    if part == 'version' and i + 1 < len(parts):
+                        self._ffmpeg_version = parts[i + 1]
+                        return self._ffmpeg_version
+            self._ffmpeg_version = 'unknown'
+            return self._ffmpeg_version
+        except Exception:
+            self._ffmpeg_version = 'unknown'
+            return self._ffmpeg_version
+
+    def _get_codec_info(self) -> str:
+        """获取支持的编解码器信息"""
+        if self._codec_info is not None:
+            return self._codec_info
+        try:
+            result = subprocess.run(
+                ['ffmpeg', '-codecs'],
+                capture_output=True,
+                text=True,
+                timeout=5
+            )
+            # 检查常用编解码器
+            codecs = []
+            output = result.stdout
+            if 'libx264' in output:
+                codecs.append('libx264')
+            if 'libx265' in output or 'hevc' in output:
+                codecs.append('libx265')
+            if 'aac' in output:
+                codecs.append('aac')
+            if 'libfdk_aac' in output:
+                codecs.append('libfdk_aac')
+            self._codec_info = ', '.join(codecs) if codecs else 'unknown'
+            return self._codec_info
+        except Exception:
+            self._codec_info = 'unknown'
+            return self._codec_info
+
+    def _get_system_info(self) -> Dict[str, Any]:
+        """获取系统信息"""
+        try:
+            now = time.monotonic()
+            if (
+                self._system_info_cache
+                and now - self._system_info_cache_ts < self.SYSTEM_INFO_TTL_SECONDS
+            ):
+                return self._system_info_cache
+
+            import platform
+            import psutil
+
+            if self._hw_accel_info is None:
+                self._hw_accel_info = get_hw_accel_info_str()
+
+            if self._static_system_info is None:
+                self._static_system_info = {
+                    'os': platform.system(),
+                    'cpu': f"{psutil.cpu_count()} cores",
+                    'memory': f"{psutil.virtual_memory().total // (1024**3)}GB",
+                    'hwAccelConfig': self.config.hw_accel,  # 当前配置的硬件加速
+                    'hwAccelSupport': self._hw_accel_info,  # 系统支持的硬件加速
+                }
+
+            info = dict(self._static_system_info)
+            info.update({
+                'cpuUsage': f"{psutil.cpu_percent()}%",
+                'memoryAvailable': f"{psutil.virtual_memory().available // (1024**3)}GB",
+            })
+
+            # 尝试获取 GPU 信息
+            gpu_info = self._get_gpu_info()
+            if gpu_info:
+                info['gpu'] = gpu_info
+
+            self._system_info_cache = info
+            self._system_info_cache_ts = now
+            return info
+        except Exception:
+            return {}
+
+    def _get_gpu_info(self) -> Optional[str]:
+        """获取 GPU 信息"""
+        if self._gpu_info_checked:
+            return self._gpu_info
+
+        self._gpu_info_checked = True
+        try:
+            result = subprocess.run(
+                ['nvidia-smi', '--query-gpu=name', '--format=csv,noheader'],
+                capture_output=True,
+                text=True,
+                timeout=5
+            )
+            if result.returncode == 0:
+                gpu_name = result.stdout.strip().split('\n')[0]
+                self._gpu_info = gpu_name
+        except Exception:
+            self._gpu_info = None
+
+        return self._gpu_info
+
+    def close(self):
+        """关闭会话"""
+        self.session.close()

services/cache.py

@@ -0,0 +1,348 @@
+# -*- coding: utf-8 -*-
+"""
+素材缓存服务
+
+提供素材下载缓存功能，避免相同素材重复下载。
+"""
+
+import os
+import hashlib
+import logging
+import shutil
+import time
+import uuid
+from typing import Optional, Tuple
+from urllib.parse import urlparse, unquote
+
+from services import storage
+
+logger = logging.getLogger(__name__)
+
+
+def _extract_cache_key(url: str) -> str:
+    """
+    从 URL 提取缓存键
+
+    去除签名等查询参数，保留路径作为唯一标识。
+
+    Args:
+        url: 完整的素材 URL
+
+    Returns:
+        缓存键（URL 路径的 MD5 哈希）
+    """
+    parsed = urlparse(url)
+    # 使用 scheme + host + path 作为唯一标识（忽略签名等查询参数）
+    cache_key_source = f"{parsed.scheme}://{parsed.netloc}{unquote(parsed.path)}"
+    return hashlib.md5(cache_key_source.encode('utf-8')).hexdigest()
+
+
+def _get_file_extension(url: str) -> str:
+    """
+    从 URL 提取文件扩展名
+
+    Args:
+        url: 素材 URL
+
+    Returns:
+        文件扩展名（如 .mp4, .png），无法识别时返回空字符串
+    """
+    parsed = urlparse(url)
+    path = unquote(parsed.path)
+    _, ext = os.path.splitext(path)
+    return ext.lower() if ext else ''
+
+
+class MaterialCache:
+    """
+    素材缓存管理器
+
+    负责素材文件的缓存存储和检索。
+    """
+
+    LOCK_TIMEOUT_SEC = 30.0
+    LOCK_POLL_INTERVAL_SEC = 0.1
+
+    def __init__(self, cache_dir: str, enabled: bool = True, max_size_gb: float = 0):
+        """
+        初始化缓存管理器
+
+        Args:
+            cache_dir: 缓存目录路径
+            enabled: 是否启用缓存
+            max_size_gb: 最大缓存大小（GB），0 表示不限制
+        """
+        self.cache_dir = cache_dir
+        self.enabled = enabled
+        self.max_size_bytes = int(max_size_gb * 1024 * 1024 * 1024) if max_size_gb > 0 else 0
+
+        if self.enabled:
+            os.makedirs(self.cache_dir, exist_ok=True)
+            logger.info(f"Material cache initialized: {cache_dir}")
+
+    def get_cache_path(self, url: str) -> str:
+        """
+        获取素材的缓存文件路径
+
+        Args:
+            url: 素材 URL
+
+        Returns:
+            缓存文件的完整路径
+        """
+        cache_key = _extract_cache_key(url)
+        ext = _get_file_extension(url)
+        filename = f"{cache_key}{ext}"
+        return os.path.join(self.cache_dir, filename)
+
+    def _get_lock_path(self, cache_key: str) -> str:
+        """获取缓存锁文件路径"""
+        assert self.cache_dir
+        return os.path.join(self.cache_dir, f"{cache_key}.lock")
+
+    def _acquire_lock(self, cache_key: str) -> Optional[str]:
+        """获取缓存锁（跨进程安全）"""
+        if not self.enabled:
+            return None
+
+        lock_path = self._get_lock_path(cache_key)
+        deadline = time.monotonic() + self.LOCK_TIMEOUT_SEC
+
+        while True:
+            try:
+                fd = os.open(lock_path, os.O_CREAT | os.O_EXCL | os.O_WRONLY)
+                os.close(fd)
+                return lock_path
+            except FileExistsError:
+                if time.monotonic() >= deadline:
+                    logger.warning(f"Cache lock timeout: {lock_path}")
+                    return None
+                time.sleep(self.LOCK_POLL_INTERVAL_SEC)
+            except Exception as e:
+                logger.warning(f"Cache lock error: {e}")
+                return None
+
+    def _release_lock(self, lock_path: Optional[str]) -> None:
+        """释放缓存锁"""
+        if not lock_path:
+            return
+        try:
+            os.remove(lock_path)
+        except FileNotFoundError:
+            return
+        except Exception as e:
+            logger.warning(f"Cache lock release error: {e}")
+
+    def is_cached(self, url: str) -> Tuple[bool, str]:
+        """
+        检查素材是否已缓存
+
+        Args:
+            url: 素材 URL
+
+        Returns:
+            (是否已缓存, 缓存文件路径)
+        """
+        if not self.enabled:
+            return False, ''
+
+        cache_path = self.get_cache_path(url)
+        exists = os.path.exists(cache_path) and os.path.getsize(cache_path) > 0
+        return exists, cache_path
+
+    def get_or_download(
+        self,
+        url: str,
+        dest: str,
+        timeout: int = 300,
+        max_retries: int = 5
+    ) -> bool:
+        """
+        从缓存获取素材，若未缓存则下载并缓存
+
+        Args:
+            url: 素材 URL
+            dest: 目标文件路径（任务工作目录中的路径）
+            timeout: 下载超时时间（秒）
+            max_retries: 最大重试次数
+
+        Returns:
+            是否成功
+        """
+        # 确保目标目录存在
+        dest_dir = os.path.dirname(dest)
+        if dest_dir:
+            os.makedirs(dest_dir, exist_ok=True)
+
+        # 缓存未启用时直接下载
+        if not self.enabled:
+            return storage.download_file(url, dest, max_retries=max_retries, timeout=timeout)
+
+        cache_key = _extract_cache_key(url)
+        lock_path = self._acquire_lock(cache_key)
+        if not lock_path:
+            logger.warning(f"Cache lock unavailable, downloading without cache: {url[:80]}...")
+            return storage.download_file(url, dest, max_retries=max_retries, timeout=timeout)
+
+        try:
+            cache_path = self.get_cache_path(url)
+            cached = os.path.exists(cache_path) and os.path.getsize(cache_path) > 0
+
+            if cached:
+                # 命中缓存，复制到目标路径
+                try:
+                    shutil.copy2(cache_path, dest)
+                    # 更新访问时间（用于 LRU 清理）
+                    os.utime(cache_path, None)
+                    file_size = os.path.getsize(dest)
+                    logger.info(f"Cache hit: {url[:80]}... -> {dest} ({file_size} bytes)")
+                    return True
+                except Exception as e:
+                    logger.warning(f"Failed to copy from cache: {e}, will re-download")
+                    # 缓存复制失败，删除可能损坏的缓存文件
+                    try:
+                        os.remove(cache_path)
+                    except Exception:
+                        pass
+
+            # 未命中缓存，下载到缓存目录
+            logger.debug(f"Cache miss: {url[:80]}...")
+
+            # 先下载到临时文件（唯一文件名，避免并发覆盖）
+            temp_cache_path = os.path.join(
+                self.cache_dir,
+                f"{cache_key}.{uuid.uuid4().hex}.downloading"
+            )
+            try:
+                if not storage.download_file(url, temp_cache_path, max_retries=max_retries, timeout=timeout):
+                    # 下载失败，清理临时文件
+                    if os.path.exists(temp_cache_path):
+                        os.remove(temp_cache_path)
+                    return False
+
+                if not os.path.exists(temp_cache_path) or os.path.getsize(temp_cache_path) <= 0:
+                    if os.path.exists(temp_cache_path):
+                        os.remove(temp_cache_path)
+                    return False
+
+                # 下载成功，原子替换缓存文件
+                os.replace(temp_cache_path, cache_path)
+
+                # 复制到目标路径
+                shutil.copy2(cache_path, dest)
+                file_size = os.path.getsize(dest)
+                logger.info(f"Downloaded and cached: {url[:80]}... ({file_size} bytes)")
+
+                # 检查是否需要清理缓存
+                if self.max_size_bytes > 0:
+                    self._cleanup_if_needed()
+
+                return True
+
+            except Exception as e:
+                logger.error(f"Cache download error: {e}")
+                # 清理临时文件
+                if os.path.exists(temp_cache_path):
+                    try:
+                        os.remove(temp_cache_path)
+                    except Exception:
+                        pass
+                return False
+        finally:
+            self._release_lock(lock_path)
+
+    def _cleanup_if_needed(self) -> None:
+        """
+        检查并清理缓存（LRU 策略）
+
+        当缓存大小超过限制时，删除最久未访问的文件。
+        """
+        if self.max_size_bytes <= 0:
+            return
+
+        try:
+            # 获取所有缓存文件及其信息
+            cache_files = []
+            total_size = 0
+
+            for filename in os.listdir(self.cache_dir):
+                if filename.endswith('.downloading') or filename.endswith('.lock'):
+                    continue
+                file_path = os.path.join(self.cache_dir, filename)
+                if os.path.isfile(file_path):
+                    stat = os.stat(file_path)
+                    cache_files.append({
+                        'path': file_path,
+                        'size': stat.st_size,
+                        'atime': stat.st_atime
+                    })
+                    total_size += stat.st_size
+
+            # 如果未超过限制，无需清理
+            if total_size <= self.max_size_bytes:
+                return
+
+            # 按访问时间排序（最久未访问的在前）
+            cache_files.sort(key=lambda x: x['atime'])
+
+            # 删除文件直到低于限制的 80%
+            target_size = int(self.max_size_bytes * 0.8)
+            deleted_count = 0
+
+            for file_info in cache_files:
+                if total_size <= target_size:
+                    break
+                try:
+                    os.remove(file_info['path'])
+                    total_size -= file_info['size']
+                    deleted_count += 1
+                except Exception as e:
+                    logger.warning(f"Failed to delete cache file: {e}")
+
+            if deleted_count > 0:
+                logger.info(f"Cache cleanup: deleted {deleted_count} files, current size: {total_size / (1024*1024*1024):.2f} GB")
+
+        except Exception as e:
+            logger.warning(f"Cache cleanup error: {e}")
+
+    def clear(self) -> None:
+        """清空所有缓存"""
+        if not self.enabled:
+            return
+
+        try:
+            if os.path.exists(self.cache_dir):
+                shutil.rmtree(self.cache_dir)
+                os.makedirs(self.cache_dir, exist_ok=True)
+                logger.info("Cache cleared")
+        except Exception as e:
+            logger.error(f"Failed to clear cache: {e}")
+
+    def get_stats(self) -> dict:
+        """
+        获取缓存统计信息
+
+        Returns:
+            包含缓存统计的字典
+        """
+        if not self.enabled or not os.path.exists(self.cache_dir):
+            return {'enabled': False, 'file_count': 0, 'total_size_mb': 0}
+
+        file_count = 0
+        total_size = 0
+
+        for filename in os.listdir(self.cache_dir):
+            if filename.endswith('.downloading') or filename.endswith('.lock'):
+                continue
+            file_path = os.path.join(self.cache_dir, filename)
+            if os.path.isfile(file_path):
+                file_count += 1
+                total_size += os.path.getsize(file_path)
+
+        return {
+            'enabled': True,
+            'cache_dir': self.cache_dir,
+            'file_count': file_count,
+            'total_size_mb': round(total_size / (1024 * 1024), 2),
+            'max_size_gb': self.max_size_bytes / (1024 * 1024 * 1024) if self.max_size_bytes > 0 else 0
+        }

services/gpu_scheduler.py

@@ -0,0 +1,164 @@
+# -*- coding: utf-8 -*-
+"""
+GPU 调度器
+
+提供多 GPU 设备的轮询调度功能。
+"""
+
+import logging
+import threading
+from typing import List, Optional
+
+from domain.config import WorkerConfig
+from domain.gpu import GPUDevice
+from util.system import get_all_gpu_info, validate_gpu_device
+from constant import HW_ACCEL_CUDA, HW_ACCEL_QSV
+
+logger = logging.getLogger(__name__)
+
+
+class GPUScheduler:
+    """
+    GPU 调度器
+
+    实现多 GPU 设备的轮询（Round Robin）调度。
+    线程安全，支持并发任务执行。
+
+    使用方式：
+        scheduler = GPUScheduler(config)
+
+        # 在任务执行时
+        device_index = scheduler.acquire()
+        try:
+            # 执行任务
+            pass
+        finally:
+            scheduler.release(device_index)
+    """
+
+    def __init__(self, config: WorkerConfig):
+        """
+        初始化调度器
+
+        Args:
+            config: Worker 配置
+        """
+        self._config = config
+        self._devices: List[GPUDevice] = []
+        self._next_index: int = 0
+        self._lock = threading.Lock()
+        self._enabled = False
+
+        # 初始化设备列表
+        self._init_devices()
+
+    def _init_devices(self) -> None:
+        """初始化 GPU 设备列表"""
+        # 仅在启用硬件加速时才初始化
+        if self._config.hw_accel not in (HW_ACCEL_CUDA, HW_ACCEL_QSV):
+            logger.info("Hardware acceleration not enabled, GPU scheduler disabled")
+            return
+
+        configured_devices = self._config.gpu_devices
+
+        if configured_devices:
+            # 使用配置指定的设备
+            self._devices = self._validate_configured_devices(configured_devices)
+        else:
+            # 自动检测所有设备
+            self._devices = self._auto_detect_devices()
+
+        if self._devices:
+            self._enabled = True
+            device_info = ', '.join(str(d) for d in self._devices)
+            logger.info(f"GPU scheduler initialized with {len(self._devices)} device(s): {device_info}")
+        else:
+            logger.warning("No GPU devices available, scheduler disabled")
+
+    def _validate_configured_devices(self, indices: List[int]) -> List[GPUDevice]:
+        """
+        验证配置的设备列表
+
+        Args:
+            indices: 配置的设备索引列表
+
+        Returns:
+            验证通过的设备列表
+        """
+        devices = []
+        for index in indices:
+            if validate_gpu_device(index):
+                devices.append(GPUDevice(
+                    index=index,
+                    name=f"GPU-{index}",
+                    available=True
+                ))
+            else:
+                logger.warning(f"GPU device {index} is not available, skipping")
+        return devices
+
+    def _auto_detect_devices(self) -> List[GPUDevice]:
+        """
+        自动检测所有可用 GPU
+
+        Returns:
+            检测到的设备列表
+        """
+        all_devices = get_all_gpu_info()
+        # 过滤不可用设备
+        return [d for d in all_devices if d.available]
+
+    @property
+    def enabled(self) -> bool:
+        """调度器是否启用"""
+        return self._enabled
+
+    @property
+    def device_count(self) -> int:
+        """设备数量"""
+        return len(self._devices)
+
+    def acquire(self) -> Optional[int]:
+        """
+        获取下一个可用的 GPU 设备（轮询调度）
+
+        Returns:
+            GPU 设备索引，如果调度器未启用或无设备则返回 None
+        """
+        if not self._enabled or not self._devices:
+            return None
+
+        with self._lock:
+            device = self._devices[self._next_index]
+            self._next_index = (self._next_index + 1) % len(self._devices)
+            logger.debug(f"Acquired GPU device: {device.index}")
+            return device.index
+
+    def release(self, device_index: Optional[int]) -> None:
+        """
+        释放 GPU 设备
+
+        当前实现为无状态轮询，此方法仅用于日志记录。
+
+        Args:
+            device_index: 设备索引
+        """
+        if device_index is not None:
+            logger.debug(f"Released GPU device: {device_index}")
+
+    def get_status(self) -> dict:
+        """
+        获取调度器状态信息
+
+        Returns:
+            状态字典
+        """
+        return {
+            'enabled': self._enabled,
+            'device_count': len(self._devices),
+            'devices': [
+                {'index': d.index, 'name': d.name, 'available': d.available}
+                for d in self._devices
+            ],
+            'hw_accel': self._config.hw_accel,
+        }

services/lease_service.py

@@ -0,0 +1,110 @@
+# -*- coding: utf-8 -*-
+"""
+租约续期服务
+
+后台线程定期为正在执行的任务续期租约。
+"""
+
+import logging
+import threading
+import time
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    from services.api_client import APIClientV2
+
+logger = logging.getLogger(__name__)
+
+
+class LeaseService:
+    """
+    租约续期服务
+
+    在后台线程中定期调用 API 延长任务租约，
+    防止长时间任务因租约过期被回收。
+    """
+
+    def __init__(
+        self,
+        api_client: 'APIClientV2',
+        task_id: str,
+        interval: int = 60,
+        extension: int = 300
+    ):
+        """
+        初始化租约服务
+
+        Args:
+            api_client: API 客户端
+            task_id: 任务 ID
+            interval: 续期间隔（秒），默认 60 秒
+            extension: 每次续期时长（秒），默认 300 秒
+        """
+        self.api_client = api_client
+        self.task_id = task_id
+        self.interval = interval
+        self.extension = extension
+        self.running = False
+        self.thread: threading.Thread = None
+        self._stop_event = threading.Event()
+
+    def start(self):
+        """启动租约续期线程"""
+        if self.running:
+            logger.warning(f"[task:{self.task_id}] Lease service already running")
+            return
+
+        self.running = True
+        self._stop_event.clear()
+        self.thread = threading.Thread(
+            target=self._run,
+            name=f"LeaseService-{self.task_id}",
+            daemon=True
+        )
+        self.thread.start()
+        logger.debug(f"[task:{self.task_id}] Lease service started (interval={self.interval}s)")
+
+    def stop(self):
+        """停止租约续期线程"""
+        if not self.running:
+            return
+
+        self.running = False
+        self._stop_event.set()
+
+        if self.thread and self.thread.is_alive():
+            self.thread.join(timeout=5)
+
+        logger.debug(f"[task:{self.task_id}] Lease service stopped")
+
+    def _run(self):
+        """续期线程主循环"""
+        while self.running:
+            # 等待指定间隔或收到停止信号
+            if self._stop_event.wait(timeout=self.interval):
+                # 收到停止信号
+                break
+
+            if self.running:
+                self._extend_lease()
+
+    def _extend_lease(self):
+        """执行租约续期"""
+        try:
+            success = self.api_client.extend_lease(self.task_id, self.extension)
+            if success:
+                logger.debug(f"[task:{self.task_id}] Lease extended by {self.extension}s")
+            else:
+                logger.warning(f"[task:{self.task_id}] Failed to extend lease")
+        except Exception as e:
+            logger.warning(f"[task:{self.task_id}] Lease extension error: {e}")
+
+    def __enter__(self):
+        """上下文管理器入口"""
+        self.start()
+        return self
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        """上下文管理器出口"""
+        self.stop()
+        return False

services/storage.py

@@ -0,0 +1,236 @@
+# -*- coding: utf-8 -*-
+"""
+存储服务
+
+提供文件上传/下载功能，支持 OSS 签名 URL 和 HTTP_REPLACE_MAP 环境变量。
+"""
+
+import os
+import logging
+import subprocess
+from typing import Optional
+
+import requests
+
+logger = logging.getLogger(__name__)
+
+
+# 文件扩展名到 Content-Type 的映射
+_CONTENT_TYPE_MAP = {
+    '.mp4': 'video/mp4',
+    '.aac': 'audio/aac',
+    '.ts': 'video/mp2t',
+    '.m4a': 'audio/mp4',
+}
+
+
+def _get_content_type(file_path: str) -> str:
+    """
+    根据文件扩展名获取 Content-Type
+
+    Args:
+        file_path: 文件路径
+
+    Returns:
+        Content-Type 字符串
+    """
+    ext = os.path.splitext(file_path)[1].lower()
+    return _CONTENT_TYPE_MAP.get(ext, 'application/octet-stream')
+
+
+def _apply_http_replace_map(url: str) -> str:
+    """
+    应用 HTTP_REPLACE_MAP 环境变量替换 URL
+
+    Args:
+        url: 原始 URL
+
+    Returns:
+        替换后的 URL
+    """
+    replace_map = os.getenv("HTTP_REPLACE_MAP", "")
+    if not replace_map:
+        return url
+
+    new_url = url
+    replace_list = [i.split("|", 1) for i in replace_map.split(",") if "|" in i]
+    for src, dst in replace_list:
+        new_url = new_url.replace(src, dst)
+
+    if new_url != url:
+        logger.debug(f"HTTP_REPLACE_MAP: {url} -> {new_url}")
+
+    return new_url
+
+
+def upload_file(url: str, file_path: str, max_retries: int = 5, timeout: int = 60) -> bool:
+    """
+    使用签名 URL 上传文件到 OSS
+
+    Args:
+        url: 签名 URL
+        file_path: 本地文件路径
+        max_retries: 最大重试次数
+        timeout: 超时时间（秒）
+
+    Returns:
+        是否成功
+    """
+    if not os.path.exists(file_path):
+        logger.error(f"File not found: {file_path}")
+        return False
+
+    file_size = os.path.getsize(file_path)
+    logger.info(f"Uploading: {file_path} ({file_size} bytes)")
+
+    # 检查是否使用 rclone 上传
+    if os.getenv("UPLOAD_METHOD") == "rclone":
+        result = _upload_with_rclone(url, file_path)
+        if result:
+            return True
+        # rclone 失败时回退到 HTTP
+
+    # 应用 HTTP_REPLACE_MAP 替换 URL
+    http_url = _apply_http_replace_map(url)
+    content_type = _get_content_type(file_path)
+    logger.info(f"Uploading to: {http_url} (Content-Type: {content_type})")
+
+    retries = 0
+    while retries < max_retries:
+        try:
+            with open(file_path, 'rb') as f:
+                with requests.put(
+                    http_url,
+                    data=f,
+                    stream=True,
+                    timeout=timeout,
+                    headers={"Content-Type": content_type}
+                ) as response:
+                    response.raise_for_status()
+                    logger.info(f"Upload succeeded: {file_path}")
+                    return True
+
+        except requests.exceptions.Timeout:
+            retries += 1
+            logger.warning(f"Upload timed out. Retrying {retries}/{max_retries}...")
+        except requests.exceptions.RequestException as e:
+            retries += 1
+            logger.warning(f"Upload failed ({e}). Retrying {retries}/{max_retries}...")
+
+    logger.error(f"Upload failed after {max_retries} retries: {file_path}")
+    return False
+
+
+def _upload_with_rclone(url: str, file_path: str) -> bool:
+    """
+    使用 rclone 上传文件
+
+    Args:
+        url: 目标 URL
+        file_path: 本地文件路径
+
+    Returns:
+        是否成功
+    """
+    replace_map = os.getenv("RCLONE_REPLACE_MAP", "")
+    if not replace_map:
+        return False
+
+    config_file = os.getenv("RCLONE_CONFIG_FILE", "")
+
+    # 替换 URL
+    new_url = url
+    replace_list = [i.split("|", 1) for i in replace_map.split(",") if "|" in i]
+    for src, dst in replace_list:
+        new_url = new_url.replace(src, dst)
+    new_url = new_url.split("?", 1)[0]  # 移除查询参数
+
+    if new_url == url:
+        return False
+
+    cmd = [
+        "rclone",
+        "copyto",
+        "--no-check-dest",
+        "--ignore-existing",
+        "--multi-thread-chunk-size",
+        "8M",
+        "--multi-thread-streams",
+        "8",
+    ]
+    if config_file:
+        cmd.extend(["--config", config_file])
+    cmd.extend([file_path, new_url])
+
+    logger.debug(f"rclone command: {' '.join(cmd)}")
+
+    result = subprocess.run(cmd, capture_output=True, text=True)
+    if result.returncode == 0:
+        logger.info(f"rclone upload succeeded: {file_path}")
+        return True
+
+    stderr = (result.stderr or '').strip()
+    stderr = stderr[:500] if stderr else ""
+    logger.warning(f"rclone upload failed (code={result.returncode}): {file_path} {stderr}")
+    return False
+
+
+def download_file(
+    url: str,
+    file_path: str,
+    max_retries: int = 5,
+    timeout: int = 30,
+    skip_if_exist: bool = False
+) -> bool:
+    """
+    使用签名 URL 下载文件
+
+    Args:
+        url: 签名 URL
+        file_path: 本地文件路径
+        max_retries: 最大重试次数
+        timeout: 超时时间（秒）
+        skip_if_exist: 如果文件存在则跳过
+
+    Returns:
+        是否成功
+    """
+    # 如果文件已存在且跳过
+    if skip_if_exist and os.path.exists(file_path):
+        logger.debug(f"File exists, skipping download: {file_path}")
+        return True
+
+    logger.info(f"Downloading: {url}")
+
+    # 确保目标目录存在
+    file_dir = os.path.dirname(file_path)
+    if file_dir:
+        os.makedirs(file_dir, exist_ok=True)
+
+    # 应用 HTTP_REPLACE_MAP 替换 URL
+    http_url = _apply_http_replace_map(url)
+
+    retries = 0
+    while retries < max_retries:
+        try:
+            with requests.get(http_url, timeout=timeout, stream=True) as response:
+                response.raise_for_status()
+
+                with open(file_path, 'wb') as f:
+                    for chunk in response.iter_content(chunk_size=8192):
+                        if chunk:
+                            f.write(chunk)
+
+            file_size = os.path.getsize(file_path)
+            logger.info(f"Download succeeded: {file_path} ({file_size} bytes)")
+            return True
+
+        except requests.exceptions.Timeout:
+            retries += 1
+            logger.warning(f"Download timed out. Retrying {retries}/{max_retries}...")
+        except requests.exceptions.RequestException as e:
+            retries += 1
+            logger.warning(f"Download failed ({e}). Retrying {retries}/{max_retries}...")
+
+    logger.error(f"Download failed after {max_retries} retries: {url}")
+    return False

services/task_executor.py

@@ -0,0 +1,278 @@
+# -*- coding: utf-8 -*-
+"""
+任务执行器
+
+管理任务的并发执行，协调处理器、租约服务等组件。
+"""
+
+import logging
+import threading
+from concurrent.futures import ThreadPoolExecutor, Future
+from typing import Dict, Optional, TYPE_CHECKING
+
+from domain.task import Task, TaskType
+from domain.result import TaskResult, ErrorCode
+
+# 需要 GPU 加速的任务类型
+GPU_REQUIRED_TASK_TYPES = {
+    TaskType.RENDER_SEGMENT_VIDEO,
+    TaskType.COMPOSE_TRANSITION,
+}
+from domain.config import WorkerConfig
+from core.handler import TaskHandler
+from services.lease_service import LeaseService
+from services.gpu_scheduler import GPUScheduler
+
+if TYPE_CHECKING:
+    from services.api_client import APIClientV2
+
+logger = logging.getLogger(__name__)
+
+
+class TaskExecutor:
+    """
+    任务执行器
+
+    负责任务的并发调度和执行，包括：
+    - 注册和管理任务处理器
+    - 维护任务执行状态
+    - 协调租约续期
+    - 上报执行结果
+    """
+
+    def __init__(self, config: WorkerConfig, api_client: 'APIClientV2'):
+        """
+        初始化任务执行器
+
+        Args:
+            config: Worker 配置
+            api_client: API 客户端
+        """
+        self.config = config
+        self.api_client = api_client
+
+        # 任务处理器注册表
+        self.handlers: Dict[TaskType, TaskHandler] = {}
+
+        # 当前任务跟踪
+        self.current_tasks: Dict[str, Task] = {}
+        self.current_futures: Dict[str, Future] = {}
+
+        # 线程池
+        self.executor = ThreadPoolExecutor(
+            max_workers=config.max_concurrency,
+            thread_name_prefix="TaskWorker"
+        )
+
+        # 线程安全锁
+        self.lock = threading.Lock()
+
+        # GPU 调度器（如果启用硬件加速）
+        self.gpu_scheduler = GPUScheduler(config)
+
+        if self.gpu_scheduler.enabled:
+            logger.info(f"GPU scheduler enabled with {self.gpu_scheduler.device_count} device(s)")
+
+        # 注册处理器
+        self._register_handlers()
+
+    def _register_handlers(self):
+        """注册所有任务处理器"""
+        # 延迟导入以避免循环依赖
+        from handlers.render_video import RenderSegmentVideoHandler
+        from handlers.compose_transition import ComposeTransitionHandler
+        from handlers.prepare_audio import PrepareJobAudioHandler
+        from handlers.package_ts import PackageSegmentTsHandler
+        from handlers.finalize_mp4 import FinalizeMp4Handler
+
+        handlers = [
+            RenderSegmentVideoHandler(self.config, self.api_client),
+            ComposeTransitionHandler(self.config, self.api_client),
+            PrepareJobAudioHandler(self.config, self.api_client),
+            PackageSegmentTsHandler(self.config, self.api_client),
+            FinalizeMp4Handler(self.config, self.api_client),
+        ]
+
+        for handler in handlers:
+            task_type = handler.get_supported_type()
+            self.handlers[task_type] = handler
+            logger.debug(f"Registered handler for {task_type.value}")
+
+    def get_current_task_ids(self) -> list:
+        """
+        获取当前正在执行的任务 ID 列表
+
+        Returns:
+            任务 ID 列表
+        """
+        with self.lock:
+            return list(self.current_tasks.keys())
+
+    def get_current_task_count(self) -> int:
+        """
+        获取当前正在执行的任务数量
+
+        Returns:
+            任务数量
+        """
+        with self.lock:
+            return len(self.current_tasks)
+
+    def can_accept_task(self) -> bool:
+        """
+        检查是否可以接受新任务
+
+        Returns:
+            是否可以接受
+        """
+        return self.get_current_task_count() < self.config.max_concurrency
+
+    def submit_task(self, task: Task) -> bool:
+        """
+        提交任务到线程池
+
+        Args:
+            task: 任务实体
+
+        Returns:
+            是否提交成功
+        """
+        with self.lock:
+            # 检查任务是否已在执行
+            if task.task_id in self.current_tasks:
+                logger.warning(f"[task:{task.task_id}] Task already running, skipping")
+                return False
+
+            # 检查并发上限
+            if len(self.current_tasks) >= self.config.max_concurrency:
+                logger.info(
+                    f"[task:{task.task_id}] Max concurrency reached "
+                    f"({self.config.max_concurrency}), rejecting task"
+                )
+                return False
+
+            # 检查是否有对应的处理器
+            if task.task_type not in self.handlers:
+                logger.error(f"[task:{task.task_id}] No handler for type: {task.task_type.value}")
+                return False
+
+            # 记录任务
+            self.current_tasks[task.task_id] = task
+
+            # 提交到线程池
+            future = self.executor.submit(self._process_task, task)
+            self.current_futures[task.task_id] = future
+
+            logger.info(f"[task:{task.task_id}] Submitted ({task.task_type.value})")
+            return True
+
+    def _process_task(self, task: Task):
+        """
+        处理单个任务（在线程池中执行）
+
+        Args:
+            task: 任务实体
+        """
+        task_id = task.task_id
+        logger.info(f"[task:{task_id}] Starting {task.task_type.value}")
+
+        # 启动租约续期服务
+        lease_service = LeaseService(
+            self.api_client,
+            task_id,
+            interval=self.config.lease_extension_threshold,
+            extension=self.config.lease_extension_duration
+        )
+        lease_service.start()
+
+        # 获取 GPU 设备（仅对需要 GPU 的任务类型）
+        device_index = None
+        needs_gpu = task.task_type in GPU_REQUIRED_TASK_TYPES
+        if needs_gpu and self.gpu_scheduler.enabled:
+            device_index = self.gpu_scheduler.acquire()
+            if device_index is not None:
+                logger.info(f"[task:{task_id}] Assigned to GPU device {device_index}")
+
+        # 获取处理器（需要在设置 GPU 设备前获取）
+        handler = self.handlers.get(task.task_type)
+
+        try:
+            # 报告任务开始
+            self.api_client.report_start(task_id)
+
+            if not handler:
+                raise ValueError(f"No handler for task type: {task.task_type}")
+
+            # 设置 GPU 设备（线程本地存储）
+            if device_index is not None:
+                handler.set_gpu_device(device_index)
+
+            # 执行前钩子
+            handler.before_handle(task)
+
+            # 执行任务
+            result = handler.handle(task)
+
+            # 执行后钩子
+            handler.after_handle(task, result)
+
+            # 上报结果
+            if result.success:
+                self.api_client.report_success(task_id, result.data)
+                logger.info(f"[task:{task_id}] Completed successfully")
+            else:
+                error_code = result.error_code.value if result.error_code else 'E_UNKNOWN'
+                self.api_client.report_fail(task_id, error_code, result.error_message or '')
+                logger.error(f"[task:{task_id}] Failed: {result.error_message}")
+
+        except Exception as e:
+            logger.error(f"[task:{task_id}] Exception: {e}", exc_info=True)
+            self.api_client.report_fail(task_id, 'E_UNKNOWN', str(e))
+
+        finally:
+            # 清除 GPU 设备设置
+            if handler:
+                handler.clear_gpu_device()
+
+            # 释放 GPU 设备（仅当实际分配了设备时）
+            if device_index is not None:
+                self.gpu_scheduler.release(device_index)
+
+            # 停止租约续期
+            lease_service.stop()
+
+            # 从当前任务中移除
+            with self.lock:
+                self.current_tasks.pop(task_id, None)
+                self.current_futures.pop(task_id, None)
+
+    def shutdown(self, wait: bool = True):
+        """
+        关闭执行器
+
+        Args:
+            wait: 是否等待所有任务完成
+        """
+        logger.info("Shutting down task executor...")
+
+        # 关闭线程池
+        self.executor.shutdown(wait=wait)
+
+        # 清理状态
+        with self.lock:
+            self.current_tasks.clear()
+            self.current_futures.clear()
+
+        logger.info("Task executor shutdown complete")
+
+    def get_handler(self, task_type: TaskType) -> Optional[TaskHandler]:
+        """
+        获取指定类型的处理器
+
+        Args:
+            task_type: 任务类型
+
+        Returns:
+            处理器实例，不存在则返回 None
+        """
+        return self.handlers.get(task_type)

telemetry/__init__.py

@@ -1,37 +0,0 @@
-import os
-
-from constant import SOFTWARE_VERSION
-from opentelemetry import trace
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter as OTLPSpanHttpExporter
-from opentelemetry.sdk.resources import DEPLOYMENT_ENVIRONMENT, HOST_NAME, Resource, SERVICE_NAME, SERVICE_VERSION
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import BatchSpanProcessor, SimpleSpanProcessor
-from opentelemetry.instrumentation.threading import ThreadingInstrumentor
-
-ThreadingInstrumentor().instrument()
-
-def get_tracer(name):
-    return trace.get_tracer(name)
-
-# 初始化 OpenTelemetry
-def init_opentelemetry(batch=True):
-    # 设置服务名、主机名
-    resource = Resource(attributes={
-        SERVICE_NAME: "RENDER_WORKER",
-        SERVICE_VERSION: SOFTWARE_VERSION,
-        DEPLOYMENT_ENVIRONMENT: "Python",
-        HOST_NAME: os.getenv("ACCESS_KEY"),
-    })
-
-    # 使用HTTP协议上报
-    if batch:
-        span_processor = BatchSpanProcessor(OTLPSpanHttpExporter(
-            endpoint="https://oltp.jerryyan.top/v1/traces",
-        ))
-    else:
-        span_processor = SimpleSpanProcessor(OTLPSpanHttpExporter(
-            endpoint="https://oltp.jerryyan.top/v1/traces",
-        ))
-
-    trace_provider = TracerProvider(resource=resource, active_span_processor=span_processor)
-    trace.set_tracer_provider(trace_provider)

template/.gitignore

@@ -1 +0,0 @@
-**/*

template/__init__.py

@@ -1,128 +0,0 @@
-import json
-import os
-import logging
-
-from telemetry import get_tracer
-from util import api, oss
-
-TEMPLATES = {}
-logger = logging.getLogger("template")
-
-def check_local_template(local_name):
-    template_def = TEMPLATES[local_name]
-    base_dir = template_def.get("local_path")
-    for video_part in template_def.get("video_parts", []):
-        source_file = video_part.get("source", "")
-        if str(source_file).startswith("http"):
-            # download file
-            ...
-        elif str(source_file).startswith("PLACEHOLDER_"):
-            continue
-        else:
-            if not os.path.isabs(source_file):
-                source_file = os.path.join(base_dir, source_file)
-            if not os.path.exists(source_file):
-                logger.error(f"{source_file} not found, please check the template definition")
-                raise Exception(f"{source_file} not found, please check the template definition")
-        for audio in video_part.get("audios", []):
-            if not os.path.isabs(audio):
-                audio = os.path.join(base_dir, audio)
-            if not os.path.exists(audio):
-                logger.error(f"{audio} not found, please check the template definition")
-                raise Exception(f"{audio} not found, please check the template definition")
-        for lut in video_part.get("luts", []):
-            if not os.path.isabs(lut):
-                lut = os.path.join(base_dir, lut)
-            if not os.path.exists(lut):
-                logger.error(f"{lut} not found, please check the template definition")
-                raise Exception(f"{lut} not found, please check the template definition")
-        for mask in video_part.get("overlays", []):
-            if not os.path.isabs(mask):
-                mask = os.path.join(base_dir, mask)
-            if not os.path.exists(mask):
-                logger.error(f"{mask} not found, please check the template definition")
-                raise Exception(f"{mask} not found, please check the template definition")
-
-
-def load_template(template_name, local_path):
-    global TEMPLATES
-    logger.info(f"加载视频模板定义：【{template_name}（{local_path}）】")
-    template_def_file = os.path.join(local_path, "template.json")
-    if os.path.exists(template_def_file):
-        TEMPLATES[template_name] = json.load(open(template_def_file, 'rb'))
-        TEMPLATES[template_name]["local_path"] = local_path
-        try:
-            check_local_template(template_name)
-            logger.info(f"完成加载【{template_name}】模板")
-        except Exception as e:
-            logger.error(f"模板定义文件【{template_def_file}】有误，正在尝试重新下载模板", exc_info=e)
-            download_template(template_name)
-
-
-def load_local_template():
-    for template_name in os.listdir(os.getenv("TEMPLATE_DIR")):
-        if template_name.startswith("_"):
-            continue
-        if template_name.startswith("."):
-            continue
-        target_path = os.path.join(os.getenv("TEMPLATE_DIR"), template_name)
-        if os.path.isdir(target_path):
-            load_template(template_name, target_path)
-
-
-def get_template_def(template_id):
-    if template_id not in TEMPLATES:
-        download_template(template_id)
-    return TEMPLATES.get(template_id)
-
-def download_template(template_id):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("download_template"):
-        template_info = api.get_template_info(template_id)
-        if template_info is None:
-            return
-        if not os.path.isdir(template_info['local_path']):
-            os.makedirs(template_info['local_path'])
-        # download template assets
-        overall_template = template_info['overall_template']
-        video_parts = template_info['video_parts']
-        def _download_assets(_template):
-            if 'source' in _template:
-                if str(_template['source']).startswith("http"):
-                    _, _fn = os.path.split(_template['source'])
-                    new_fp = os.path.join(template_info['local_path'], _fn)
-                    oss.download_from_oss(_template['source'], new_fp)
-                    if _fn.endswith(".mp4"):
-                        from util.ffmpeg import re_encode_and_annexb
-                        new_fp = re_encode_and_annexb(new_fp)
-                    _template['source'] = os.path.relpath(new_fp, template_info['local_path'])
-            if 'overlays' in _template:
-                for i in range(len(_template['overlays'])):
-                    overlay = _template['overlays'][i]
-                    if str(overlay).startswith("http"):
-                        _, _fn = os.path.split(overlay)
-                        oss.download_from_oss(overlay, os.path.join(template_info['local_path'], _fn))
-                        _template['overlays'][i] = _fn
-            if 'luts' in _template:
-                for i in range(len(_template['luts'])):
-                    lut = _template['luts'][i]
-                    if str(lut).startswith("http"):
-                        _, _fn = os.path.split(lut)
-                        oss.download_from_oss(lut, os.path.join(template_info['local_path'], _fn))
-                        _template['luts'][i] = _fn
-            if 'audios' in _template:
-                for i in range(len(_template['audios'])):
-                    if str(_template['audios'][i]).startswith("http"):
-                        _, _fn = os.path.split(_template['audios'][i])
-                        oss.download_from_oss(_template['audios'][i], os.path.join(template_info['local_path'], _fn))
-                        _template['audios'][i] = _fn
-        _download_assets(overall_template)
-        for video_part in video_parts:
-            _download_assets(video_part)
-        with open(os.path.join(template_info['local_path'], 'template.json'), 'w', encoding='utf-8') as f:
-            json.dump(template_info, f)
-        load_template(template_id, template_info['local_path'])
-
-
-def analyze_template(template_id):
-    ...

tests/unit/test_material_cache_lock.py

@@ -0,0 +1,15 @@
+# -*- coding: utf-8 -*-
+
+import os
+
+from services.cache import MaterialCache, _extract_cache_key
+
+
+def test_cache_lock_acquire_release(tmp_path):
+    cache = MaterialCache(cache_dir=str(tmp_path), enabled=True, max_size_gb=0)
+    cache_key = _extract_cache_key("https://example.com/path/file.mp4?token=abc")
+    lock_path = cache._acquire_lock(cache_key)
+    assert lock_path
+    assert os.path.exists(lock_path)
+    cache._release_lock(lock_path)
+    assert not os.path.exists(lock_path)

util/__init__.py

@@ -0,0 +1,15 @@
+# -*- coding: utf-8 -*-
+"""
+工具模块
+
+提供系统信息采集等工具函数。
+"""
+
+from util.system import get_sys_info, get_capabilities, get_gpu_info, get_ffmpeg_version
+
+__all__ = [
+    'get_sys_info',
+    'get_capabilities',
+    'get_gpu_info',
+    'get_ffmpeg_version',
+]

util/api.py

@@ -1,260 +0,0 @@
-import json
-import logging
-import os
-import threading
-
-import requests
-from opentelemetry.trace import Status, StatusCode
-
-import util.system
-from telemetry import get_tracer
-from util import oss
-
-session = requests.Session()
-logger = logging.getLogger(__name__)
-
-
-def normalize_task(task_info):
-    ...
-    return task_info
-
-
-def sync_center():
-    """
-    通过接口获取任务
-    :return: 任务列表
-    """
-    from template import TEMPLATES, download_template
-    try:
-        response = session.post(os.getenv('API_ENDPOINT') + "/sync", json={
-            'accessKey': os.getenv('ACCESS_KEY'),
-            'clientStatus': util.system.get_sys_info(),
-            'templateList': [{'id': t.get('id', ''), 'updateTime': t.get('updateTime', '')} for t in
-                             TEMPLATES.values()]
-        }, timeout=10)
-        response.raise_for_status()
-    except requests.RequestException as e:
-        logger.error("请求失败！", e)
-        return []
-    data = response.json()
-    logger.debug("获取任务结果：【%s】", data)
-    if data.get('code', 0) == 200:
-        templates = data.get('data', {}).get('templates', [])
-        tasks = data.get('data', {}).get('tasks', [])
-    else:
-        tasks = []
-        templates = []
-        logger.warning("获取任务失败")
-    if os.getenv("REDIRECT_TO_URL", False) != False:
-        for task in tasks:
-            _sess = requests.Session()
-            logger.info("重定向任务【%s】至配置的地址：%s", task.get("id"), os.getenv("REDIRECT_TO_URL"))
-            url = f"{os.getenv('REDIRECT_TO_URL')}{task.get('id')}"
-            threading.Thread(target=requests.post, args=(url,)).start()
-        return []
-    for template in templates:
-        template_id = template.get('id', '')
-        if template_id:
-            logger.info("更新模板：【%s】", template_id)
-            download_template(template_id)
-    return tasks
-
-
-def get_template_info(template_id):
-    """
-    通过接口获取模板信息
-    :rtype: Template
-    :param template_id: 模板id
-    :type template_id: str
-    :return: 模板信息
-    """
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("get_template_info"):
-        with tracer.start_as_current_span("get_template_info.request") as req_span:
-            try:
-                req_span.set_attribute("http.method", "POST")
-                req_span.set_attribute("http.url", '{0}/template/{1}'.format(os.getenv('API_ENDPOINT'), template_id))
-                response = session.post('{0}/template/{1}'.format(os.getenv('API_ENDPOINT'), template_id), json={
-                    'accessKey': os.getenv('ACCESS_KEY'),
-                }, timeout=10)
-                req_span.set_attribute("http.status_code", response.status_code)
-                req_span.set_attribute("http.response", response.text)
-                response.raise_for_status()
-            except requests.RequestException as e:
-                req_span.set_attribute("api.error", str(e))
-                logger.error("请求失败！", e)
-                return None
-        data = response.json()
-        logger.debug("获取模板信息结果：【%s】", data)
-        remote_template_info = data.get('data', {})
-        if not remote_template_info:
-            logger.warning("获取模板信息结果为空", data)
-            return None
-        template = {
-            'id': template_id,
-            'updateTime': remote_template_info.get('updateTime', template_id),
-            'scenic_name': remote_template_info.get('scenicName', '景区'),
-            'name': remote_template_info.get('name', '模版'),
-            'video_size': remote_template_info.get('resolution', '1920x1080'),
-            'frame_rate': 25,
-            'overall_duration': 30,
-            'video_parts': [
-
-            ]
-        }
-
-        def _template_normalizer(template_info):
-            _template = {}
-            _placeholder_type = template_info.get('isPlaceholder', -1)
-            if _placeholder_type == 0:
-                # 固定视频
-                _template['source'] = template_info.get('sourceUrl', '')
-            elif _placeholder_type == 1:
-                # 占位符
-                _template['source'] = "PLACEHOLDER_" + template_info.get('sourceUrl', '')
-                _template['mute'] = template_info.get('mute', True)
-                _template['crop_mode'] = template_info.get('cropEnable', None)
-                _template['zoom_cut'] = template_info.get('zoomCut', None)
-            else:
-                _template['source'] = None
-            _overlays = template_info.get('overlays', '')
-            if _overlays:
-                _template['overlays'] = _overlays.split(",")
-            _audios = template_info.get('audios', '')
-            if _audios:
-                _template['audios'] = _audios.split(",")
-            _luts = template_info.get('luts', '')
-            if _luts:
-                _template['luts'] = _luts.split(",")
-            _only_if = template_info.get('onlyIf', '')
-            if _only_if:
-                _template['only_if'] = _only_if
-            _effects = template_info.get('effects', '')
-            if _effects:
-                _template['effects'] = _effects.split("|")
-            return _template
-
-        # outer template definition
-        overall_template = _template_normalizer(remote_template_info)
-        template['overall_template'] = overall_template
-        # inter template definition
-        inter_template_list = remote_template_info.get('children', [])
-        for children_template in inter_template_list:
-            parts = _template_normalizer(children_template)
-            template['video_parts'].append(parts)
-        template['local_path'] = os.path.join(os.getenv('TEMPLATE_DIR'), str(template_id))
-        with get_tracer("api").start_as_current_span("get_template_info.template") as res_span:
-            res_span.set_attribute("normalized.response", json.dumps(template))
-        return template
-
-
-def report_task_success(task_info, **kwargs):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("report_task_success"):
-        with tracer.start_as_current_span("report_task_success.request") as req_span:
-            try:
-                req_span.set_attribute("http.method", "POST")
-                req_span.set_attribute("http.url",
-                                       '{0}/{1}/success'.format(os.getenv('API_ENDPOINT'), task_info.get("id")))
-                response = session.post('{0}/{1}/success'.format(os.getenv('API_ENDPOINT'), task_info.get("id")), json={
-                    'accessKey': os.getenv('ACCESS_KEY'),
-                    **kwargs
-                }, timeout=10)
-                req_span.set_attribute("http.status_code", response.status_code)
-                req_span.set_attribute("http.response", response.text)
-                response.raise_for_status()
-                req_span.set_status(Status(StatusCode.OK))
-            except requests.RequestException as e:
-                req_span.set_attribute("api.error", str(e))
-                logger.error("请求失败！", e)
-                return None
-
-
-def report_task_start(task_info):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("report_task_start"):
-        with tracer.start_as_current_span("report_task_start.request") as req_span:
-            try:
-                req_span.set_attribute("http.method", "POST")
-                req_span.set_attribute("http.url",
-                                       '{0}/{1}/start'.format(os.getenv('API_ENDPOINT'), task_info.get("id")))
-                response = session.post('{0}/{1}/start'.format(os.getenv('API_ENDPOINT'), task_info.get("id")), json={
-                    'accessKey': os.getenv('ACCESS_KEY'),
-                }, timeout=10)
-                req_span.set_attribute("http.status_code", response.status_code)
-                req_span.set_attribute("http.response", response.text)
-                response.raise_for_status()
-                req_span.set_status(Status(StatusCode.OK))
-            except requests.RequestException as e:
-                req_span.set_attribute("api.error", str(e))
-                logger.error("请求失败！", e)
-                return None
-
-
-def report_task_failed(task_info, reason=''):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("report_task_failed") as span:
-        span.set_attribute("task_id", task_info.get("id"))
-        span.set_attribute("reason", reason)
-        with tracer.start_as_current_span("report_task_failed.request") as req_span:
-            try:
-                req_span.set_attribute("http.method", "POST")
-                req_span.set_attribute("http.url",
-                                       '{0}/{1}/fail'.format(os.getenv('API_ENDPOINT'), task_info.get("id")))
-                response = session.post('{0}/{1}/fail'.format(os.getenv('API_ENDPOINT'), task_info.get("id")), json={
-                    'accessKey': os.getenv('ACCESS_KEY'),
-                    'reason': reason
-                }, timeout=10)
-                req_span.set_attribute("http.status_code", response.status_code)
-                req_span.set_attribute("http.response", response.text)
-                response.raise_for_status()
-                req_span.set_status(Status(StatusCode.OK))
-            except requests.RequestException as e:
-                req_span.set_attribute("api.error", str(e))
-                req_span.set_status(Status(StatusCode.ERROR))
-                logger.error("请求失败！", e)
-                return None
-
-
-def upload_task_file(task_info, ffmpeg_task):
-    tracer = get_tracer(__name__)
-    with get_tracer("api").start_as_current_span("upload_task_file") as span:
-        logger.info("开始上传文件: %s", task_info.get("id"))
-        span.set_attribute("file.id", task_info.get("id"))
-        with tracer.start_as_current_span("upload_task_file.request_upload_url") as req_span:
-            try:
-                req_span.set_attribute("http.method", "POST")
-                req_span.set_attribute("http.url",
-                                       '{0}/{1}/uploadUrl'.format(os.getenv('API_ENDPOINT'), task_info.get("id")))
-                response = session.post('{0}/{1}/uploadUrl'.format(os.getenv('API_ENDPOINT'), task_info.get("id")),
-                                        json={
-                                            'accessKey': os.getenv('ACCESS_KEY'),
-                                        }, timeout=10)
-                req_span.set_attribute("http.status_code", response.status_code)
-                req_span.set_attribute("http.response", response.text)
-                response.raise_for_status()
-                req_span.set_status(Status(StatusCode.OK))
-            except requests.RequestException as e:
-                span.set_attribute("api.error", str(e))
-                req_span.set_status(Status(StatusCode.ERROR))
-                logger.error("请求失败！", e)
-                return False
-        data = response.json()
-        url = data.get('data', "")
-        logger.info("开始上传文件: %s 至 %s", task_info.get("id"), url)
-        return oss.upload_to_oss(url, ffmpeg_task.get_output_file())
-
-
-def get_task_info(id):
-    try:
-        response = session.get(os.getenv('API_ENDPOINT') + "/" + id + "/info", params={
-            'accessKey': os.getenv('ACCESS_KEY'),
-        }, timeout=10)
-        response.raise_for_status()
-    except requests.RequestException as e:
-        logger.error("请求失败！", e)
-        return []
-    data = response.json()
-    logger.debug("获取任务结果：【%s】", data)
-    if data.get('code', 0) == 200:
-        return data.get('data', {})

util/ffmpeg.py

@@ -1,255 +0,0 @@
-import json
-import logging
-import os
-import subprocess
-from datetime import datetime
-from typing import Optional, IO
-
-from opentelemetry.trace import Status, StatusCode
-
-from entity.ffmpeg import FfmpegTask, ENCODER_ARGS, VIDEO_ARGS, AUDIO_ARGS, MUTE_AUDIO_INPUT, get_mp4toannexb_filter
-from telemetry import get_tracer
-
-logger = logging.getLogger(__name__)
-
-def re_encode_and_annexb(file):
-    with get_tracer("ffmpeg").start_as_current_span("re_encode_and_annexb") as span:
-        span.set_attribute("file.path", file)
-        if not os.path.exists(file):
-            span.set_status(Status(StatusCode.ERROR))
-            return file
-        logger.info("ReEncodeAndAnnexb: %s", file)
-        has_audio = not not probe_video_audio(file)
-        # 优先使用RE_ENCODE_VIDEO_ARGS环境变量，其次使用默认的VIDEO_ARGS
-        if os.getenv("RE_ENCODE_VIDEO_ARGS", False):
-            _video_args = tuple(os.getenv("RE_ENCODE_VIDEO_ARGS", "").split(" "))
-        else:
-            _video_args = VIDEO_ARGS
-        # 优先使用RE_ENCODE_ENCODER_ARGS环境变量，其次使用默认的ENCODER_ARGS
-        if os.getenv("RE_ENCODE_ENCODER_ARGS", False):
-            _encoder_args = tuple(os.getenv("RE_ENCODE_ENCODER_ARGS", "").split(" "))
-        else:
-            _encoder_args = ENCODER_ARGS
-        ffmpeg_process = subprocess.run(["ffmpeg", "-y", "-hide_banner", "-i", file,
-                                         *(set() if has_audio else MUTE_AUDIO_INPUT),
-                                         "-fps_mode", "cfr",
-                                         "-map", "0:v", "-map", "0:a" if has_audio else "1:a",
-                                         *_video_args, "-bsf:v", get_mp4toannexb_filter(),
-                                         *AUDIO_ARGS, "-bsf:a", "setts=pts=DTS",
-                                         *_encoder_args, "-shortest", "-fflags", "+genpts",
-                                         "-f", "mpegts", file + ".ts"])
-        logger.info(" ".join(ffmpeg_process.args))
-        span.set_attribute("ffmpeg.args", json.dumps(ffmpeg_process.args))
-        logger.info("ReEncodeAndAnnexb: %s, returned: %s", file, ffmpeg_process.returncode)
-        span.set_attribute("ffmpeg.code", ffmpeg_process.returncode)
-        if ffmpeg_process.returncode == 0:
-            span.set_status(Status(StatusCode.OK))
-            span.set_attribute("file.size", os.path.getsize(file+".ts"))
-            # os.remove(file)
-            return file+".ts"
-        else:
-            span.set_status(Status(StatusCode.ERROR))
-            return file
-
-def start_render(ffmpeg_task: FfmpegTask):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("start_render") as span:
-        span.set_attribute("ffmpeg.task", str(ffmpeg_task))
-        if not ffmpeg_task.need_run():
-            ffmpeg_task.set_output_file(ffmpeg_task.input_file[0])
-            span.set_status(Status(StatusCode.OK))
-            return True
-        ffmpeg_args = ffmpeg_task.get_ffmpeg_args()
-        if len(ffmpeg_args) == 0:
-            ffmpeg_task.set_output_file(ffmpeg_task.input_file[0])
-            span.set_status(Status(StatusCode.OK))
-            return True
-        ffmpeg_process = subprocess.run(["ffmpeg", "-progress", "-", "-loglevel", "error", *ffmpeg_args], stderr=subprocess.PIPE, **subprocess_args(True))
-        span.set_attribute("ffmpeg.args", json.dumps(ffmpeg_process.args))
-        logger.info(" ".join(ffmpeg_process.args))
-        ffmpeg_final_out = handle_ffmpeg_output(ffmpeg_process.stdout)
-        span.set_attribute("ffmpeg.out", ffmpeg_final_out)
-        logger.info("FINISH TASK, OUTPUT IS %s", ffmpeg_final_out)
-        code = ffmpeg_process.returncode
-        span.set_attribute("ffmpeg.code", code)
-        if code != 0:
-            span.set_attribute("ffmpeg.err", str(ffmpeg_process.stderr))
-            span.set_status(Status(StatusCode.ERROR, "FFMPEG异常退出"))
-            logger.error("FFMPEG ERROR: %s", ffmpeg_process.stderr)
-            return False
-        span.set_attribute("ffmpeg.out_file", ffmpeg_task.output_file)
-        try:
-            file_size = os.path.getsize(ffmpeg_task.output_file)
-            span.set_attribute("file.size", file_size)
-            if file_size < 4096:
-                span.set_status(Status(StatusCode.ERROR, "输出文件过小"))
-                logger.error("FFMPEG ERROR: OUTPUT FILE IS TOO SMALL")
-                return False
-        except OSError as e:
-            span.set_attribute("file.size", 0)
-            span.set_attribute("file.error", e.strerror)
-            span.set_status(Status(StatusCode.ERROR, "输出文件不存在"))
-            logger.error("FFMPEG ERROR: OUTPUT FILE NOT FOUND")
-            return False
-        span.set_status(Status(StatusCode.OK))
-        return True
-
-def handle_ffmpeg_output(stdout: Optional[bytes]) -> str:
-    out_time = "0:0:0.0"
-    if stdout is None:
-        print("[!]STDOUT is null")
-        return out_time
-    speed = "0"
-    for line in stdout.split(b"\n"):
-        if line == b"":
-            break
-        if line.strip() == b"progress=end":
-            # 处理完毕
-            break
-        if line.startswith(b"out_time="):
-            out_time = line.replace(b"out_time=", b"").decode().strip()
-        if line.startswith(b"speed="):
-            speed = line.replace(b"speed=", b"").decode().strip()
-    print("[ ]Speed:", out_time, "@", speed)
-    return out_time+"@"+speed
-
-def duration_str_to_float(duration_str: str) -> float:
-    _duration = datetime.strptime(duration_str, "%H:%M:%S.%f") - datetime(1900, 1, 1)
-    return _duration.total_seconds()
-
-
-def probe_video_info(video_file):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("probe_video_info") as span:
-        span.set_attribute("video.file", video_file)
-        # 获取宽度和高度
-        result = subprocess.run(
-            ["ffprobe", '-v', 'error', '-select_streams', 'v:0', '-show_entries', 'stream=width,height:format=duration', '-of',
-             'csv=s=x:p=0', video_file],
-            stderr=subprocess.STDOUT,
-            **subprocess_args(True)
-        )
-        span.set_attribute("ffprobe.args", json.dumps(result.args))
-        span.set_attribute("ffprobe.code", result.returncode)
-        if result.returncode != 0:
-            span.set_status(Status(StatusCode.ERROR))
-            return 0, 0, 0
-        all_result = result.stdout.decode('utf-8').strip()
-        span.set_attribute("ffprobe.out", all_result)
-        if all_result == '':
-            span.set_status(Status(StatusCode.ERROR))
-            return 0, 0, 0
-        span.set_status(Status(StatusCode.OK))
-        wh, duration = all_result.split('\n')
-        width, height = wh.strip().split('x')
-        return int(width), int(height), float(duration)
-
-
-def probe_video_audio(video_file, type=None):
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("probe_video_audio") as span:
-        span.set_attribute("video.file", video_file)
-        args = ["ffprobe", "-hide_banner", "-v", "error", "-select_streams", "a", "-show_entries", "stream=index", "-of", "csv=p=0"]
-        if type == 'concat':
-            args.append("-safe")
-            args.append("0")
-            args.append("-f")
-            args.append("concat")
-        args.append(video_file)
-        logger.info(" ".join(args))
-        result = subprocess.run(args, stderr=subprocess.STDOUT, **subprocess_args(True))
-        span.set_attribute("ffprobe.args", json.dumps(result.args))
-        span.set_attribute("ffprobe.code", result.returncode)
-        logger.info("probe_video_audio: %s", result.stdout.decode('utf-8').strip())
-        if result.returncode != 0:
-            return False
-        if result.stdout.decode('utf-8').strip() == '':
-            return False
-        return True
-
-
-# 音频淡出2秒
-def fade_out_audio(file, duration, fade_out_sec = 2):
-    if type(duration) == str:
-        try:
-            duration = float(duration)
-        except Exception as e:
-            logger.error("duration is not float: %s", e)
-            return file
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("fade_out_audio") as span:
-        span.set_attribute("audio.file", file)
-        if duration <= fade_out_sec:
-            return file
-        else:
-            new_fn = file + "_.mp4"
-            if os.path.exists(new_fn):
-                os.remove(new_fn)
-                logger.info("delete tmp file: " + new_fn)
-            try:
-                process = subprocess.run(["ffmpeg", "-i", file, "-c:v", "copy", "-c:a", "aac", "-af", "afade=t=out:st=" + str(duration - fade_out_sec) + ":d=" + str(fade_out_sec), "-y", new_fn], **subprocess_args(True))
-                span.set_attribute("ffmpeg.args", json.dumps(process.args))
-                logger.info(" ".join(process.args))
-                if process.returncode != 0:
-                    span.set_status(Status(StatusCode.ERROR))
-                    logger.error("FFMPEG ERROR: %s", process.stderr)
-                    return file
-                else:
-                    span.set_status(Status(StatusCode.OK))
-                    return new_fn
-            except Exception as e:
-                span.set_status(Status(StatusCode.ERROR))
-                logger.error("FFMPEG ERROR: %s", e)
-                return file
-
-
-
-# Create a set of arguments which make a ``subprocess.Popen`` (and
-# variants) call work with or without Pyinstaller, ``--noconsole`` or
-# not, on Windows and Linux. Typical use::
-#
-#   subprocess.call(['program_to_run', 'arg_1'], **subprocess_args())
-#
-# When calling ``check_output``::
-#
-#   subprocess.check_output(['program_to_run', 'arg_1'],
-#                           **subprocess_args(False))
-def subprocess_args(include_stdout=True):
-    # The following is true only on Windows.
-    if hasattr(subprocess, 'STARTUPINFO'):
-        # On Windows, subprocess calls will pop up a command window by default
-        # when run from Pyinstaller with the ``--noconsole`` option. Avoid this
-        # distraction.
-        si = subprocess.STARTUPINFO()
-        si.dwFlags |= subprocess.STARTF_USESHOWWINDOW
-        # Windows doesn't search the path by default. Pass it an environment so
-        # it will.
-        env = os.environ
-    else:
-        si = None
-        env = None
-
-    # ``subprocess.check_output`` doesn't allow specifying ``stdout``::
-    #
-    #   Traceback (most recent call last):
-    #     File "test_subprocess.py", line 58, in <module>
-    #       **subprocess_args(stdout=None))
-    #     File "C:\Python27\lib\subprocess.py", line 567, in check_output
-    #       raise ValueError('stdout argument not allowed, it will be overridden.')
-    #   ValueError: stdout argument not allowed, it will be overridden.
-    #
-    # So, add it only if it's needed.
-    if include_stdout:
-        ret = {'stdout': subprocess.PIPE}
-    else:
-        ret = {}
-
-    # On Windows, running this from the binary produced by Pyinstaller
-    # with the ``--noconsole`` option requires redirecting everything
-    # (stdin, stdout, stderr) to avoid an OSError exception
-    # "[Error 6] the handle is invalid."
-    ret.update({'stdin': subprocess.PIPE,
-                'startupinfo': si,
-                'env': env})
-    return ret
-

util/oss.py

@@ -1,127 +0,0 @@
-import logging
-import os
-import sys
-
-import requests
-from opentelemetry.trace import Status, StatusCode
-
-from telemetry import get_tracer
-
-logger = logging.getLogger(__name__)
-
-
-def upload_to_oss(url, file_path):
-    """
-    使用签名URL上传文件到OSS
-    :param str url: 签名URL
-    :param str file_path: 文件路径
-    :return bool: 是否成功
-    """
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("upload_to_oss") as span:
-        span.set_attribute("file.url", url)
-        span.set_attribute("file.path", file_path)
-        span.set_attribute("file.size", os.path.getsize(file_path))
-        max_retries = 5
-        retries = 0
-        if os.getenv("UPLOAD_METHOD") == "rclone":
-            with tracer.start_as_current_span("rclone_to_oss") as r_span:
-                replace_map = os.getenv("RCLONE_REPLACE_MAP")
-                r_span.set_attribute("rclone.replace_map", replace_map)
-                if replace_map != "":
-                    replace_list = [i.split("|", 1) for i in replace_map.split(",")]
-                    new_url = url
-                    for (_src, _dst) in replace_list:
-                        new_url = new_url.replace(_src, _dst)
-                    new_url = new_url.split("?", 1)[0]
-                    r_span.set_attribute("rclone.target_dir", new_url)
-                    if new_url != url:
-                        result = os.system(f"rclone copyto --no-check-dest --ignore-existing --multi-thread-chunk-size 8M --multi-thread-streams 8 {file_path} {new_url}")
-                        r_span.set_attribute("rclone.result", result)
-                        if result == 0:
-                            span.set_status(Status(StatusCode.OK))
-                            return True
-                        else:
-                            span.set_status(Status(StatusCode.ERROR))
-        while retries < max_retries:
-            with tracer.start_as_current_span("upload_to_oss.request") as req_span:
-                req_span.set_attribute("http.retry_count", retries)
-                try:
-                    req_span.set_attribute("http.method", "PUT")
-                    req_span.set_attribute("http.url", url)
-                    with open(file_path, 'rb') as f:
-                        response = requests.put(url, data=f, stream=True, timeout=60, headers={"Content-Type": "video/mp4"})
-                        req_span.set_attribute("http.status_code", response.status_code)
-                        req_span.set_attribute("http.response", response.text)
-                        response.raise_for_status()
-                        req_span.set_status(Status(StatusCode.OK))
-                        span.set_status(Status(StatusCode.OK))
-                        return True
-                except requests.exceptions.Timeout:
-                    req_span.set_attribute("http.error", "Timeout")
-                    req_span.set_status(Status(StatusCode.ERROR))
-                    retries += 1
-                    logger.warning(f"Upload timed out. Retrying {retries}/{max_retries}...")
-                except Exception as e:
-                    req_span.set_attribute("http.error", str(e))
-                    req_span.set_status(Status(StatusCode.ERROR))
-                    retries += 1
-                    logger.warning(f"Upload failed. Retrying {retries}/{max_retries}...")
-        span.set_status(Status(StatusCode.ERROR))
-    return False
-
-
-def download_from_oss(url, file_path, skip_if_exist=None):
-    """
-    使用签名URL下载文件到OSS
-    :param skip_if_exist: 如果存在就不下载了
-    :param str url: 签名URL
-    :param Union[LiteralString, str, bytes] file_path: 文件路径
-    :return bool: 是否成功
-    """
-    tracer = get_tracer(__name__)
-    with tracer.start_as_current_span("download_from_oss") as span:
-        span.set_attribute("file.url", url)
-        span.set_attribute("file.path", file_path)
-        
-        # 如果skip_if_exist为None，则从启动参数中读取
-        if skip_if_exist is None:
-            skip_if_exist = 'skip_if_exist' in sys.argv
-        
-        if skip_if_exist and os.path.exists(file_path):
-            span.set_attribute("file.exist", True)
-            span.set_attribute("file.size", os.path.getsize(file_path))
-            return True
-        logging.info("download_from_oss: %s", url)
-        file_dir, file_name = os.path.split(file_path)
-        if file_dir:
-            if not os.path.exists(file_dir):
-                os.makedirs(file_dir)
-        max_retries = 5
-        retries = 0
-        while retries < max_retries:
-            with tracer.start_as_current_span("download_from_oss.request") as req_span:
-                req_span.set_attribute("http.retry_count", retries)
-                try:
-                    req_span.set_attribute("http.method", "GET")
-                    req_span.set_attribute("http.url", url)
-                    response = requests.get(url, timeout=15)  # 设置超时时间
-                    req_span.set_attribute("http.status_code", response.status_code)
-                    with open(file_path, 'wb') as f:
-                        f.write(response.content)
-                    req_span.set_attribute("file.size", os.path.getsize(file_path))
-                    req_span.set_status(Status(StatusCode.OK))
-                    span.set_status(Status(StatusCode.OK))
-                    return True
-                except requests.exceptions.Timeout:
-                    req_span.set_attribute("http.error", "Timeout")
-                    req_span.set_status(Status(StatusCode.ERROR))
-                    retries += 1
-                    logger.warning(f"Download timed out. Retrying {retries}/{max_retries}...")
-                except Exception as e:
-                    req_span.set_attribute("http.error", str(e))
-                    req_span.set_status(Status(StatusCode.ERROR))
-                    retries += 1
-                    logger.warning(f"Download failed. Retrying {retries}/{max_retries}...")
-        span.set_status(Status(StatusCode.ERROR))
-    return False

util/system.py

@@ -1,24 +1,345 @@
+# -*- coding: utf-8 -*-
+"""
+系统信息工具
+
+提供系统信息采集功能。
+"""
+
+import logging
 import os
 import platform
-from datetime import datetime
+import subprocess
+from typing import Optional, Dict, Any, List
 
 import psutil
-from constant import SUPPORT_FEATURE, SOFTWARE_VERSION
+from constant import SOFTWARE_VERSION, DEFAULT_CAPABILITIES, HW_ACCEL_NONE, HW_ACCEL_QSV, HW_ACCEL_CUDA
+from domain.gpu import GPUDevice
+
+logger = logging.getLogger(__name__)
 
 
 def get_sys_info():
     """
-    Returns a dictionary with system information.
+    获取系统信息
+
+    Returns:
+        dict: 系统信息字典
     """
+    mem = psutil.virtual_memory()
+
     info = {
-        'version': SOFTWARE_VERSION,
+        'os': platform.system(),
-        'client_datetime': datetime.now().isoformat(),
+        'cpu': f"{os.cpu_count()} cores",
+        'memory': f"{mem.total // (1024**3)}GB",
+        'cpuUsage': f"{psutil.cpu_percent()}%",
+        'memoryAvailable': f"{mem.available // (1024**3)}GB",
         'platform': platform.system(),
-        'runtime_version': 'Python ' + platform.python_version(),
+        'pythonVersion': platform.python_version(),
-        'cpu_count': os.cpu_count(),
+        'version': SOFTWARE_VERSION,
-        'cpu_usage': psutil.cpu_percent(),
-        'memory_total': psutil.virtual_memory().total,
-        'memory_available': psutil.virtual_memory().available,
-        'support_feature': SUPPORT_FEATURE
     }
+
+    # 尝试获取 GPU 信息
+    gpu_info = get_gpu_info()
+    if gpu_info:
+        info['gpu'] = gpu_info
+
     return info
+
+
+def get_capabilities():
+    """
+    获取 Worker 支持的能力列表
+
+    Returns:
+        list: 能力列表
+    """
+    return DEFAULT_CAPABILITIES.copy()
+
+
+def get_gpu_info() -> Optional[str]:
+    """
+    尝试获取 GPU 信息
+
+    Returns:
+        str: GPU 信息，失败返回 None
+    """
+    try:
+        # 尝试使用 nvidia-smi
+        result = subprocess.run(
+            ['nvidia-smi', '--query-gpu=name', '--format=csv,noheader'],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        if result.returncode == 0:
+            gpu_name = result.stdout.strip().split('\n')[0]
+            return gpu_name
+    except Exception:
+        pass
+
+    return None
+
+
+def get_ffmpeg_version() -> str:
+    """
+    获取 FFmpeg 版本
+
+    Returns:
+        str: FFmpeg 版本号
+    """
+    try:
+        result = subprocess.run(
+            ['ffmpeg', '-version'],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        if result.returncode == 0:
+            first_line = result.stdout.split('\n')[0]
+            # 解析版本号，例如 "ffmpeg version 6.0 ..."
+            parts = first_line.split()
+            for i, part in enumerate(parts):
+                if part == 'version' and i + 1 < len(parts):
+                    return parts[i + 1]
+    except Exception:
+        pass
+
+    return 'unknown'
+
+
+def check_ffmpeg_encoder(encoder: str) -> bool:
+    """
+    检查 FFmpeg 是否支持指定的编码器
+
+    Args:
+        encoder: 编码器名称，如 'h264_nvenc', 'h264_qsv'
+
+    Returns:
+        bool: 是否支持该编码器
+    """
+    try:
+        result = subprocess.run(
+            ['ffmpeg', '-hide_banner', '-encoders'],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        if result.returncode == 0:
+            return encoder in result.stdout
+    except Exception:
+        pass
+
+    return False
+
+
+def check_ffmpeg_decoder(decoder: str) -> bool:
+    """
+    检查 FFmpeg 是否支持指定的解码器
+
+    Args:
+        decoder: 解码器名称，如 'h264_cuvid', 'h264_qsv'
+
+    Returns:
+        bool: 是否支持该解码器
+    """
+    try:
+        result = subprocess.run(
+            ['ffmpeg', '-hide_banner', '-decoders'],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        if result.returncode == 0:
+            return decoder in result.stdout
+    except Exception:
+        pass
+
+    return False
+
+
+def check_ffmpeg_hwaccel(hwaccel: str) -> bool:
+    """
+    检查 FFmpeg 是否支持指定的硬件加速方法
+
+    Args:
+        hwaccel: 硬件加速方法，如 'cuda', 'qsv', 'dxva2', 'd3d11va'
+
+    Returns:
+        bool: 是否支持该硬件加速方法
+    """
+    try:
+        result = subprocess.run(
+            ['ffmpeg', '-hide_banner', '-hwaccels'],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        if result.returncode == 0:
+            return hwaccel in result.stdout
+    except Exception:
+        pass
+
+    return False
+
+
+def detect_hw_accel_support() -> Dict[str, Any]:
+    """
+    检测系统的硬件加速支持情况
+
+    Returns:
+        dict: 硬件加速支持信息
+        {
+            'cuda': {
+                'available': bool,
+                'gpu': str or None,
+                'encoder': bool,  # h264_nvenc
+                'decoder': bool,  # h264_cuvid
+            },
+            'qsv': {
+                'available': bool,
+                'encoder': bool,  # h264_qsv
+                'decoder': bool,  # h264_qsv
+            },
+            'recommended': str  # 推荐的加速方式: 'cuda', 'qsv', 'none'
+        }
+    """
+    result = {
+        'cuda': {
+            'available': False,
+            'gpu': None,
+            'encoder': False,
+            'decoder': False,
+        },
+        'qsv': {
+            'available': False,
+            'encoder': False,
+            'decoder': False,
+        },
+        'recommended': HW_ACCEL_NONE
+    }
+
+    # 检测 CUDA/NVENC 支持
+    gpu_info = get_gpu_info()
+    if gpu_info:
+        result['cuda']['gpu'] = gpu_info
+        result['cuda']['available'] = check_ffmpeg_hwaccel('cuda')
+        result['cuda']['encoder'] = check_ffmpeg_encoder('h264_nvenc')
+        result['cuda']['decoder'] = check_ffmpeg_decoder('h264_cuvid')
+
+    # 检测 QSV 支持
+    result['qsv']['available'] = check_ffmpeg_hwaccel('qsv')
+    result['qsv']['encoder'] = check_ffmpeg_encoder('h264_qsv')
+    result['qsv']['decoder'] = check_ffmpeg_decoder('h264_qsv')
+
+    # 推荐硬件加速方式（优先 CUDA，其次 QSV）
+    if result['cuda']['available'] and result['cuda']['encoder']:
+        result['recommended'] = HW_ACCEL_CUDA
+    elif result['qsv']['available'] and result['qsv']['encoder']:
+        result['recommended'] = HW_ACCEL_QSV
+
+    return result
+
+
+def get_hw_accel_info_str() -> str:
+    """
+    获取硬件加速支持信息的可读字符串
+
+    Returns:
+        str: 硬件加速支持信息描述
+    """
+    support = detect_hw_accel_support()
+
+    parts = []
+
+    if support['cuda']['available']:
+        gpu = support['cuda']['gpu'] or 'Unknown GPU'
+        status = 'encoder+decoder' if support['cuda']['encoder'] and support['cuda']['decoder'] else (
+            'encoder only' if support['cuda']['encoder'] else 'decoder only' if support['cuda']['decoder'] else 'hwaccel only'
+        )
+        parts.append(f"CUDA({gpu}, {status})")
+
+    if support['qsv']['available']:
+        status = 'encoder+decoder' if support['qsv']['encoder'] and support['qsv']['decoder'] else (
+            'encoder only' if support['qsv']['encoder'] else 'decoder only' if support['qsv']['decoder'] else 'hwaccel only'
+        )
+        parts.append(f"QSV({status})")
+
+    if not parts:
+        return "No hardware acceleration available"
+
+    return ', '.join(parts) + f" [recommended: {support['recommended']}]"
+
+
+def get_all_gpu_info() -> List[GPUDevice]:
+    """
+    获取所有 NVIDIA GPU 信息
+
+    使用 nvidia-smi 查询所有 GPU 设备。
+
+    Returns:
+        GPU 设备列表，失败返回空列表
+    """
+    try:
+        result = subprocess.run(
+            [
+                'nvidia-smi',
+                '--query-gpu=index,name,memory.total',
+                '--format=csv,noheader,nounits'
+            ],
+            capture_output=True,
+            text=True,
+            timeout=10
+        )
+
+        if result.returncode != 0:
+            return []
+
+        devices = []
+        for line in result.stdout.strip().split('\n'):
+            if not line.strip():
+                continue
+            parts = [p.strip() for p in line.split(',')]
+            if len(parts) >= 2:
+                index = int(parts[0])
+                name = parts[1]
+                memory = int(parts[2]) if len(parts) >= 3 else None
+                devices.append(GPUDevice(
+                    index=index,
+                    name=name,
+                    memory_total=memory,
+                    available=True
+                ))
+
+        return devices
+
+    except Exception as e:
+        logger.warning(f"Failed to detect GPUs: {e}")
+        return []
+
+
+def validate_gpu_device(index: int) -> bool:
+    """
+    验证指定索引的 GPU 设备是否可用
+
+    Args:
+        index: GPU 设备索引
+
+    Returns:
+        设备是否可用
+    """
+    try:
+        result = subprocess.run(
+            [
+                'nvidia-smi',
+                '-i', str(index),
+                '--query-gpu=name',
+                '--format=csv,noheader'
+            ],
+            capture_output=True,
+            text=True,
+            timeout=5
+        )
+        return result.returncode == 0 and bool(result.stdout.strip())
+    except Exception:
+        return False
+