comi-logic/modalDeploy
7
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

更新2个ffmpeg worker方法和web调用接口

Browse Source
This commit is contained in:
shuohigh@gmail.com
2025-05-07 19:00:32 +08:00
parent a05e153717
commit d1557920f5
2 changed files with 235 additions and 143 deletions
Download Patch File Download Diff File Expand all files Collapse all files

318
src/cluster/ffmpeg_worker/worker.py
View File

@@ -12,10 +12,10 @@ ffmpeg_worker_image = (
app = modal.App(image=ffmpeg_worker_image, )
with ffmpeg_worker_image.imports():
from src.cluster.config import config
from typing import List, Optional, Tuple, Dict, Any
from loguru import logger
from ffmpeg.asyncio import FFmpeg
from src.cluster.config import config
from src.cluster.ffmpeg_worker.Utils.PathUtils import FileUtils
from src.cluster.video_downloader.model import (MediaSources,
MediaSource,
@@ -25,8 +25,6 @@ with ffmpeg_worker_image.imports():
from src.cluster.web.model import SentryTransactionInfo
from src.cluster.ffmpeg_worker.model import FFMpegSliceSegment
import shutil, psutil, os, backoff, json, sentry_sdk
from psutil import svmem
from psutil._common import scpufreq
from modal import current_function_call_id
sentry_sdk.init(
@@ -44,6 +42,7 @@ with ffmpeg_worker_image.imports():
s3_mount = "/mntS3"
output_path_prefix = "/mnt/outputs"
@sentry_sdk.trace
def get_cache_filepath(media: MediaSource) -> MediaCache:
"""
@@ -52,7 +51,7 @@ with ffmpeg_worker_image.imports():
:return:
"""
# 本地挂载缓存
if media.protocol == MediaProtocol.s3 and media.endpoint == config.s3_region and media.bucket == config.bucket:
if media.protocol == MediaProtocol.s3 and media.endpoint == config.s3_region and media.bucket == config.s3_bucket_name:
cache_data = MediaCache(status=MediaCacheStatus.ready, cache_filepath=f"{s3_mount}/{media.get_cdn_url()}")
else:
cache_data_json = modal_kv.get(media.cache_key)
@@ -63,7 +62,7 @@ with ffmpeg_worker_image.imports():
@sentry_sdk.trace
def capture_hardware_info() -> Tuple[scpufreq, int, svmem]:
def capture_hardware_info() -> Tuple[Any, int, Any]:
cpu_freq = psutil.cpu_freq()
cpu_count = psutil.cpu_count()
mem = psutil.virtual_memory()
@@ -71,73 +70,6 @@ with ffmpeg_worker_image.imports():
return (cpu_freq, cpu_count, mem)
@sentry_sdk.trace
async def ffmpeg_slice_process(media_source: MediaSource, media_markers: List[FFMpegSliceSegment], fn_id: str) -> \
List[str]:
cache = get_cache_filepath(media_source)
logger.info(cache.cache_filepath)
outputs: List[str] = []
ffmpeg_cmd = FFmpeg().option('y').option('hide_banner').input(cache.cache_filepath)
filter_complex: List[str] = []
for index, marker in enumerate(media_markers):
segment_out_index = f"cut{index}"
filter_complex.append(
f"[0]trim=start={marker.start.total_seconds()}:end={marker.end.total_seconds()},setpts=PTS-STARTPTS[{segment_out_index}]")
ffmpeg_cmd.option('filter_complex', ';'.join(filter_complex))
for i, marker in enumerate(media_markers):
filename = FileUtils.file_path_extend(os.path.basename(cache.cache_filepath), str(i))
output_filepath = f"{output_path_prefix}/slice/outputs/{fn_id}/{filename}"
output_filepath = output_filepath.replace('//', '/')
workdir = os.path.dirname(output_filepath)
os.makedirs(workdir, exist_ok=True)
outputs.append(output_filepath)
ffmpeg_cmd.output(output_filepath,
map=f"[cut{i}]",
reset_timestamps="1",
sc_threshold="0",
g="1",
force_key_frames="expr:gte(t,n_forced*1)",
vcodec="libx264",
crf=16,
r=30, )
@ffmpeg_cmd.on("start")
def on_start(arguments: list[str]):
try:
filter_index = arguments.index("-filter_complex")
filter_content = arguments[filter_index + 1]
arguments[filter_index + 1] = f'"{filter_content}"'
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
arguments[filter_index + 1] = filter_content
except ValueError:
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
@ffmpeg_cmd.on("progress")
def on_progress(progress):
logger.info(f"处理进度: {progress}")
@ffmpeg_cmd.on("completed")
def on_completed():
logger.info(f"FFMpeg task completed.")
@ffmpeg_cmd.on("stderr")
def on_stderr(line: str):
if line.startswith('Error'):
logger.error(line)
raise RuntimeError(line)
else:
logger.warning(line)
await ffmpeg_cmd.execute()
s3_outputs = local_copy_to_s3(outputs)
return s3_outputs
@backoff.on_exception(backoff.constant, exception=Exception, max_tries=5, max_time=60, raise_on_giveup=True)
def local_copy_to_s3(local_outputs: List[str]) -> List[str]:
s3_outputs = []
@@ -151,9 +83,10 @@ with ffmpeg_worker_image.imports():
out_s3.replace(s3_mount + '/', f"s3://{config.s3_region}/{config.s3_bucket_name}/"))
return s3_outputs
@app.function(
name="concat_medias",
cpu=12, timeout=1800, memory=(2048, 6144),
cpu=12, timeout=1800,
memory=6144,
cloud="aws",
# region='ap-northeast',
max_containers=config.ffmpeg_worker_concurrency,
@@ -198,47 +131,53 @@ with ffmpeg_worker_image.imports():
logger.info(cache.cache_filepath)
ffmpeg_cmd.input(cache.cache_filepath)
input_videos.append(cache.cache_filepath)
target_width = 1080
target_height = 1920
# video_info = await print_media_streams_info(media_paths=input_videos)
video_info = print_media_streams_info(media_paths=input_videos)
filter_complex = []
total_videos = len(input_videos)
for i in range(len(input_videos)):
info = video_info[i]
streams = info.get("streams", [])
video_stream = streams[0]
# audio_stream = streams[1]
video_duration = video_stream.get("duration")
# audio_duration = audio_stream.get("duration")
'''
# 2. 统一所有视频的格式、分辨率和帧率
for i in range(total_videos):
filter_complex.extend(
[
# 先缩放到统一分辨率，然后设置帧率和格式
f"[{i}:v]scale={target_width}:{target_height}:force_original_aspect_ratio=decrease,"
f"pad={target_width}:{target_height}:(ow-iw)/2:(oh-ih)/2,"
f"setsar=1:1," # 新增强制设置SAR
f"fps=30,format=yuv420p[v{i}]",
f"[{i}:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo[a{i}]",
]
)
'''
filter_complex.extend(
[
f"[{i}:v]trim=0:{video_duration},setpts=PTS-STARTPTS[v{i}]",
f"[{i}:a]atrim=0:{video_duration},asetpts=PTS-STARTPTS[a{i}]",
]
)
# 3. 准备处理后的视频流和音频流的连接字符串
video_streams = "".join(f"[v{i}]" for i in range(len(input_videos)))
audio_streams = "".join(f"[a{i}]" for i in range(len(input_videos)))
video_streams = "".join(f"[v{i}]" for i in range(total_videos))
audio_streams = "".join(f"[a{i}]" for i in range(total_videos))
# 4. 使用concat过滤器合并视频和音频
filter_complex.extend(
[
f"{video_streams}concat=n={len(input_videos)}:v=1:a=0[vconcated]",
f"{audio_streams}concat=n={len(input_videos)}:v=0:a=1[aconcated]",
f"{video_streams}concat=n={total_videos}:v=1:a=0[vconcated]",
f"{audio_streams}concat=n={total_videos}:v=0:a=1[aconcated]",
]
)
ffmpeg_cmd.output(
output_filepath,
{
"filter_complex": ";".join(filter_complex),
"map": ["[vconcated]", "[aconcated]"],
"vcodec": "libx264",
"crf": 16,
"r": 30,
"acodec": "aac",
"ar": 44100,
"ac": 2,
"ab": "192k",
},
)
@ffmpeg_cmd.on("start")
def on_start(arguments: list[str]):
try:
@@ -268,21 +207,6 @@ with ffmpeg_worker_image.imports():
else:
logger.warning(line)
ffmpeg_cmd.output(
output_filepath,
{
"filter_complex": ";".join(filter_complex),
"map": ["[vconcated]", "[aconcated]"],
"vcodec": "libx264",
"crf": 16,
"r": 30,
"acodec": "aac",
"ar": 44100,
"ac": 2,
"ab": "192k",
},
)
await ffmpeg_cmd.execute()
fn_id = current_function_call_id()
@@ -304,11 +228,13 @@ with ffmpeg_worker_image.imports():
with transaction:
with transaction.start_child(op='ffmpeg.concat', name='视频合并任务') as span:
sentry_sdk.set_tag('fn.id', fn_id)
span.set_data('fn.id', fn_id)
span.set_data('cpu.count', cpu_count)
span.set_data('cpu.frequency', f"{cpu_freq.current / 1000:.2f} GHz")
span.set_data('memory.available', f"{total_mb} Mb")
output_path = f"{output_path_prefix}/concat/outputs/{fn_id}/output.mp4"
os.makedirs(os.path.dirname(output_path), exist_ok=True)
await ffmpeg_process(medias, output_filepath=output_path)
s3_outputs = local_copy_to_s3([output_path])
@@ -316,8 +242,8 @@ with ffmpeg_worker_image.imports():
@app.function(
name="slice_media",
cpu=12, timeout=900, memory=(2048, 4096),
cpu=12, timeout=900,
memory=6144,
cloud="aws",
max_containers=config.ffmpeg_worker_concurrency,
volumes={
@@ -334,6 +260,73 @@ with ffmpeg_worker_image.imports():
sentry_trace: Optional[SentryTransactionInfo] = None) -> Tuple[
List[str], Optional[SentryTransactionInfo]]:
@sentry_sdk.trace
async def ffmpeg_slice_process(media_source: MediaSource, media_markers: List[FFMpegSliceSegment],
fn_id: str) -> List[str]:
cache = get_cache_filepath(media_source)
logger.info(cache.cache_filepath)
outputs: List[str] = []
ffmpeg_cmd = FFmpeg().option('y').option('hide_banner').input(cache.cache_filepath)
filter_complex: List[str] = []
for index, marker in enumerate(media_markers):
segment_out_index = f"cut{index}"
filter_complex.append(
f"[0]trim=start={marker.start.total_seconds()}:end={marker.end.total_seconds()},setpts=PTS-STARTPTS[{segment_out_index}]")
ffmpeg_cmd.option('filter_complex', ';'.join(filter_complex))
for i, marker in enumerate(media_markers):
filename = FileUtils.file_path_extend(os.path.basename(cache.cache_filepath), str(i))
output_filepath = f"{output_path_prefix}/slice/outputs/{fn_id}/{filename}"
output_filepath = output_filepath.replace('//', '/')
workdir = os.path.dirname(output_filepath)
os.makedirs(workdir, exist_ok=True)
outputs.append(output_filepath)
ffmpeg_cmd.output(output_filepath,
map=f"[cut{i}]",
reset_timestamps="1",
sc_threshold="0",
g="1",
force_key_frames="expr:gte(t,n_forced*1)",
vcodec="libx264",
acodec="aac",
crf=16,
r=30, )
@ffmpeg_cmd.on("start")
def on_start(arguments: list[str]):
try:
filter_index = arguments.index("-filter_complex")
filter_content = arguments[filter_index + 1]
arguments[filter_index + 1] = f'"{filter_content}"'
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
arguments[filter_index + 1] = filter_content
except ValueError:
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
@ffmpeg_cmd.on("progress")
def on_progress(progress):
logger.info(f"处理进度: {progress}")
@ffmpeg_cmd.on("completed")
def on_completed():
logger.info(f"FFMpeg task completed.")
@ffmpeg_cmd.on("stderr")
def on_stderr(line: str):
if line.startswith('Error'):
logger.error(line)
raise RuntimeError(line)
else:
logger.warning(line)
await ffmpeg_cmd.execute()
s3_outputs = local_copy_to_s3(outputs)
return s3_outputs
fn_id = current_function_call_id()
cpu_freq, cpu_count, mem = capture_hardware_info()
total_mb = mem.total >> 20
@@ -350,6 +343,7 @@ with ffmpeg_worker_image.imports():
x_baggage=sentry_sdk.get_baggage())
with transaction:
with transaction.start_child(op='ffmpeg.slice', name='视频切割任务') as span:
sentry_sdk.set_tag('fn.id', fn_id)
span.set_data('fn.id', fn_id)
span.set_data('cpu.count', cpu_count)
span.set_data('cpu.frequency', f"{cpu_freq.current / 1000:.2f} GHz")
@@ -357,3 +351,105 @@ with ffmpeg_worker_image.imports():
outputs = await ffmpeg_slice_process(media_source=media, media_markers=markers, fn_id=fn_id)
return outputs, sentry_trace
@app.local_entrypoint()
async def test():
from ffmpeg import FFmpeg
async def ffmpeg_process(media_sources: List[str], output_filepath: str) -> None:
ffmpeg_cmd = FFmpeg().option('y').option('hide_banner')
input_videos = []
for media_source in media_sources:
# cache = get_cache_filepath(media_source)
ffmpeg_cmd.input(media_source)
input_videos.append(media_source)
target_width = 1080
target_height = 1920
# video_info = await print_media_streams_info(media_paths=input_videos)
filter_complex = []
total_videos = len(input_videos)
# 2. 统一所有视频的格式、分辨率和帧率
for i in range(total_videos):
filter_complex.extend(
[
# 先缩放到统一分辨率，然后设置帧率和格式
f"[{i}:v]scale={target_width}:{target_height}:force_original_aspect_ratio=decrease,"
f"pad={target_width}:{target_height}:(ow-iw)/2:(oh-ih)/2,"
f"setsar=1:1," # 新增强制设置SAR
f"fps=30,format=yuv420p[v{i}]",
f"[{i}:a]aformat=sample_fmts=fltp:sample_rates=44100:channel_layouts=stereo[a{i}]",
]
)
# 3. 准备处理后的视频流和音频流的连接字符串
video_streams = "".join(f"[v{i}]" for i in range(total_videos))
audio_streams = "".join(f"[a{i}]" for i in range(total_videos))
# 4. 使用concat过滤器合并视频和音频
filter_complex.extend(
[
f"{video_streams}concat=n={total_videos}:v=1:a=0[vconcated]",
f"{audio_streams}concat=n={total_videos}:v=0:a=1[aconcated]",
]
)
ffmpeg_cmd.output(
output_filepath,
{
"filter_complex": ";".join(filter_complex),
"map": ["[vconcated]", "[aconcated]"],
"vcodec": "libx264",
"crf": 16,
"r": 30,
"acodec": "aac",
"ar": 44100,
"ac": 2,
"ab": "192k",
},
)
@ffmpeg_cmd.on("start")
def on_start(arguments: list[str]):
try:
filter_index = arguments.index("-filter_complex")
filter_content = arguments[filter_index + 1]
arguments[filter_index + 1] = f'"{filter_content}"'
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
arguments[filter_index + 1] = filter_content
except ValueError:
args = " ".join(arguments)
logger.info(f"FFmpeg command:{args}")
@ffmpeg_cmd.on("progress")
def on_progress(progress):
logger.info(f"处理进度: {progress}")
@ffmpeg_cmd.on("completed")
def on_completed():
logger.info(f"FFMpeg task completed.")
@ffmpeg_cmd.on("stderr")
def on_stderr(line: str):
if line.startswith('Error'):
logger.error(line)
raise RuntimeError(line)
else:
logger.warning(line)
ffmpeg_cmd.execute()
medias = [
"./videos/1397757910405340824_0.mp4",
"./videos/1397757910405340824_1.mp4",
"./videos/1397757910405340824_2.mp4",
"./videos/1397757910405340824_3.mp4",
"./videos/1397757910405340824_4.mp4"
]
await ffmpeg_process(medias, "./videos/output.mp4")

60
src/cluster/web/worker.py
View File

@@ -35,6 +35,7 @@ with fastapi_image.imports():
fastapi_app = modal.App(image=fastapi_image)
@fastapi_app.function(scaledown_window=60,
secrets=[
modal.Secret.from_name("cf-kv-secret", environment_name=config.environment),
@@ -153,38 +154,33 @@ with fastapi_image.imports():
:param task_id:
:return: (TaskStaus, errorReason, results, sentryTransactionInfo)
"""
try:
fn_task = modal.FunctionCall.from_id(task_id)
call_graph = fn_task.get_call_graph()
task = call_graph[0]
match task.status:
case InputStatus.PENDING:
fn_task = modal.FunctionCall.from_id(task_id)
call_graph = fn_task.get_call_graph()
task = call_graph[0]
match task.status:
case InputStatus.PENDING:
return TaskStatus.running, None, None, None
case InputStatus.SUCCESS:
try:
results, sentry_trace = fn_task.get(timeout=2)
return TaskStatus.success, None, results, sentry_trace
except modal.exception.OutputExpiredError:
return TaskStatus.expired, None, None, None
except TimeoutError:
return TaskStatus.running, None, None, None
case InputStatus.SUCCESS:
try:
results, sentry_trace = fn_task.get(timeout=2)
return TaskStatus.success, None, results, sentry_trace
except modal.exception.OutputExpiredError:
return TaskStatus.expired, None, None, None
except TimeoutError:
return TaskStatus.running, None, None, None
except Exception as e:
logger.error(e)
return TaskStatus.failed, None, None, None
case _:
error = 'FAILURE'
match task.status:
case InputStatus.INIT_FAILURE:
error = 'INIT_FAILURE'
case InputStatus.TERMINATED:
error = 'TERMINATED'
case InputStatus.TIMEOUT:
error = 'TIMEOUT'
return TaskStatus.failed, error, None, None
except Exception as e:
logger.exception(e)
raise RuntimeError("Modal task status check failed")
except Exception as e:
logger.exception(e)
raise e
case _:
error = 'FAILURE'
match task.status:
case InputStatus.INIT_FAILURE:
error = 'INIT_FAILURE'
case InputStatus.TERMINATED:
error = 'TERMINATED'
case InputStatus.TIMEOUT:
error = 'TIMEOUT'
return TaskStatus.failed, error, None, None
@web_app.get("/scalar", include_in_schema=False)
async def scalar():
@@ -360,7 +356,7 @@ with fastapi_image.imports():
if transaction:
response.headers["x-trace-id"] = transaction.x_trace_id
response.headers["x-baggage"] = transaction.x_baggage
return FFMPEGSliceTaskStatusResponse(taskId=task_id, status=TaskStatus.failed, error=reason)
return FFMPEGSliceTaskStatusResponse(taskId=task_id, status=task_status, error=reason, result=results)
@web_app.post("/ffmpeg/concat",
tags=["发起任务"],