Trying to record a video, getting an error

[antongenis]

Discussed in #598

^{Originally posted by antongenis September 29, 2021}
2021/09/29 11:26:18 I [1/0] [RTSP] [session 644619446] opened by 127.0.0.1:38766
[rtsp @ 0x5555c885b900] method SETUP failed: 461 Unsupported Transport**
2021/09/29 11:26:18 I [1/0] [RTSP] [session 158090036] opened by 127.0.0.1:38766

the command I am using to send a video:

ffmpeg -re -i ipd.mp4 -c:v libx264 -preset superfast -tune zerolatency -f rtsp -rtsp_transport tcp rtsp://SS:[email protected]:8554/stream7

---

in the yml file:

runOnPublish: ffmpeg -i rtsp://localhost:$RTSP_PORT/$RTSP_PATH -rtsp_transport tcp -c:v libx264 /mnt/video-dev/$RTSP_PATH.mp4

[antongenis]

I am seeing that the file is being recorded, but not completely, not all frames are there

the original video has 247 frames , and the one recorded -220

$ ffprobe -select_streams v -show_streams stream7.mp4 2>/dev/null | grep nb_frames | sed -e 's/nb_frames=//'
220

[aler9]

An H264 stream contains frames encoded as I-frames, B-frames and P-frames. B-frames and P-frames are partial frames, they contain only the modified portion of the stream and can't be decoded autonomously. In your case, this happens:

1. You start streaming a file, this get sent to the server (I = i-frame, P = p-frame):

   I P P P P P P P P P P P I P P P P P P P P P P P I
   

2. The server starts the runOnPublish command, but some milliseconds are passed, and this is what the ffmpeg inside the command receives:

   P P P P P I P P P P P P P P P P P I
   

   The first I-frame has been transmitted before the runOnPublish ffmpeg is able to connect, therefore it is not recorded. At the same time, the subsequent P-frames can't be recorded, since the first I-frame has not been recorded and they can't be decoded. The runOnPublish ffmpeg can record only from the 2nd I-frame onward.

This is reason behind the missing frames.

An enhancement could be adding a record function directly inside the server, that is able to record all frames without the initial delay of a runOnPublish command, but at the moment this is not implemented.