runOnPublish ffmpeg process not stopped when publisher disconnects

[die3lust1gen2]

Hi,

I don't know if this is by design or maybe a bug or something related to my stupidity regarding ffmpeg . So I will post this here.

Which version are you using?

Initially started with 0.16.4 a couple of days ago.
No change with v0.17.0

Which operating system are you using?

OS

• Linux
• Windows
• macOS

Architecture

• amd64
• arm64
• arm7
• arm6

Describe the issue

in short:

1. I publish a stream to a rtsp-simple-server instance, which uses runOnPublish to record the incoming stream with ffmpeg.
2. the publisher stops streaming
3. the recording ffmpeg process does not get killed and keeps running
4. the publisher starts to stream again, but the stream/recording gets lost
5. Killing ffmpeg or restarting the server fixes my problem

In detail:

I have a simple instance running with 1 endpoint to record an incoming stream with ffmpeg. Tried it with the go standalone binary on the system itself and within a custom docker container.

paths:

  stream_record:
    runOnPublish: ffmpeg -y -nostdin -i rtsp://127.0.0.1:$RTSP_PORT/$RTSP_PATH -c copy -f segment -segment_time 120 -reset_timestamps 1 -strftime 1 '/record/stream_%Y-%m-%d_%H-%M-%S.mp4'
    runOnPublishRestart: yes

A client (my one and only 'publisher') is publishing a stream to this endpoint with ffmpeg

ffmpeg -y -nostdin -i rtsp://source -c:v h264 -preset ultrafast -tune zerolatency -profile:v baseline -pix_fmt yuv420p -r 5 -b:v 512k -rtsp_transport tcp -f rtsp rtsp://ip:8554/stream_record

This is working so far. The stream gets recorded and segmented.

When the client stops publishing the stream the rtsp-simple-server takes notice and reports this

2021/08/09 20:15:50 I [1/1] [RTSP] [conn x.x.x.x:35248] closed speed=1.44x

But the ffmpeg process is still running and not getting killed (waited for more than 10 minutes):

root     1709255  0.0  1.4  87564 14376 pts/0    S+   20:15   0:00 ffmpeg -y -nostdin -i rtsp://127.0.0.1:8554/stream_record -c copy -f segment -segment_time 120 -reset_timestamps 1 -strftime 1 /record/stream_%Y-%m-%d_%H-%M-%S.mp4

If the client/publisher connects and starts to stream again, the recording gets stuck. The running ffmpeg is not picking up the new stream data which gets lost.

rtmp-simple-server reports the new publisher:

2021/08/09 20:21:53 I [0/1] [RTSP] [session 717400457] opened by x.x.x.x:35502

Ffmpeg shows no activity, the target file doesn't change. I have to restart rstp-simple-server or kill the ffmpeg process to be able to record again.

Killing ffmpeg gives the following log output

2021/08/09 20:30:06 I [0/1] [RTSP] [conn 127.0.0.1:52872] closed
2021/08/09 20:30:06 I [0/1] [RTSP] [session 109292930] closed
2021/08/09 20:30:06 I [0/0] [RTSP] [session 954150502] closed
2021/08/09 20:30:06 I [0/0] [RTSP] [session 511532354] closed

So from my understanding rtsp-simple-server should stop the runOnPublish ffmpeg process when the last publisher disconnects. But this does not happen. Am I missing something? Is this a bug or is my ffmpeg usage messed up?

Thanks!

[aler9]

moved here #495