How to remove a branch of tee in an active GStreamer pipeline? - gstreamer

Question

everyone
The version of GStreamer I use is 1.x. I've spent a lot of time in searching a way to delete a tee branch.
In an active pipeline, a recording bin is created as below and inserted into this pipeline by branching the tee element.
"queue ! video/x-h264, width=800, height=600, framerate=10/1, stream-format=(string)byte-stream ! h264parse ! mp4mux ! filesink location=/xxxx"
It works perfectly except that I want to dynamically delete the recording bin and get a playable mp4 file. According to some discussion and tutorial, to get a correct mp4 file , we need to handle something about EOS. After trying some methods, I always got broken mp4 files.
Does anyone have sample code written in C to show me ? I'd appreciate your help.

Answer 1

Your best bet for cases like this may be to create two processes. The first process would run the video, and half of the tee it has would deliver h264 data to the second process through whatever means.
Here are two pipelines demonstrating the concept using UDP sockets.
gst-launch-1.0 videotestsrc ! x264enc ! tee name=t ! h264parse ! avdec_h264 ! videoconvert ! ximagesink t. ! queue ! h264parse ! rtph264pay ! udpsink host=localhost port=8888
gst-launch-1.0 udpsrc port=8888 num-buffers=300 ! application/x-rtp,media=video,encoding-name=H264 ! rtph264depay ! h264parse ! mp4mux ! filesink location=/tmp/264.mp4
The trick to getting that clean mp4 is to make sure an EOS event is delivered reliably.

Answer 2

Instead of dynamically adding it you just have it in the pipeline by default, and add a probe callback at the source pad of the queue in the probe callback you have to do the trick either to pass the buffer or not (GST_PAD_PROBE_DROP drops the buffer and GST_PAD_PROBE_OK passes on the buffer to next element) so when you get an event to start/stop recoding you just need to return appropriate values. And filesink you can use multifilesink instead so as to write to different files everytime you start/stop.
Note the queue which drops the buffers needs before the mux element otherwise the file would be corrupt.
Hope that helps!

Answer 3

Finally, I came up with a solution.
Let's say that there is an active pipeline including a recording bin.
"udpsrc port=4444 caps=\"application/x-rtp, media=(string)video,
clock-rate=(int)90000, encoding-name=(string)H264 ! rtph264depay !
tee name=tp tp. ! queue ! video/x-h264, width=800, height=600,
framerate=10/1 ! decodebin ! videoconvert ! video/x-raw, format=RGBA !
autovideosink"
recording bin:
"queue ! video/x-h264, width=800, height=600, framerate=10/1,
stream-format=(string)byte-stream ! h264parse ! mp4mux ! filesink
location=/xxxx"
After a period of time, we want to stop recording and save as a mp4 file, and video media is still streaming.
First, I use a blocking probe to block the src pad of tee. In this blocking probe callback, I use an event probe to catch EOS in the sink pad of filesink and do a busy waiting.
*if EOS is catched in the event probe callback
self->isGotEOS = YES;
*busy waiting in the blocking probe callback
while (self->isGotEOS == NO) {
usleep(100000);
}
Before entering the busy waiting while loop, an EOS event is created and sent to the sink pad of recording bin.
After the busy waiting is done:
usleep(200000);
[self destory_record_elements];
I think usleep(200000) is a trick. Without it, a non-playable mp4 file is usually the result. It would seem that 200ms is long enough handling the EOS.

Answer 4

I had similar problem previously, my pipeline
videotestsrc do-timestamp="TRUE" ! videoflip method=0 ! tee name=t
t. ! queue ! videoconvert ! glupload ! glshader ! autovideosink async="FALSE"
t. ! queue ! identity drop-probability=1 ! videoconvert name=conv2 ! openh264enc ! h264parse ! avimux ! multifilesink async="FALSE" post-messages=true next-file=4
Then I just change drop-probability property on identity element
drop-probability = 1 + gst_pad_send_event(conv2_sinkpad, gst_event_new_eos()); - stop recording
drop-probability = 0 - resume recording