0

Among the fragments I streamed into KVS via its gstreamer plugin, I see that hundreds of fragments - over 55% - have 0 in the field "FragmentLengthInMilliseconds" (fetched via the LIST_FRAGMENTS api). They all have the same timestamp. Another thing I observed is that in these cases, the value of "FragmentSizeInBytes" is almost always 12265.

I wonder why these fragments even exist? Am I missing something in my gstreamer plugin configuration - to get rid of them (if they are useless)? Here is how I configured it in C++:

   g_object_set(kvs_sink,
        "stream-name", stream_name.c_str(),
        "storage-size", 1024,
        "retention-period", 168, // in hours
        "aws-region", AWS_REGION, NULL);

This is the plugin I'm using https://github.com/awslabs/amazon-kinesis-video-streams-producer-sdk-cpp

Tony Jiang
  • 442
  • 5
  • 16

1 Answers1

1

Your configuration looks OK. Fragment duration is calculated in the backend by adding the frame durations for the entire fragment. Some media pipelines have no way of specifying the frame durations and the fragment duration then evaluates to the timestamp of the last frame minus the timestamp of the first frame. In case of a single frame fragments and no duration specified, it will evaluate to 0. The 0 duration points to an issue with your media pipeline upstream. You might want to double check your configuration. If you are actually streaming video, then ensure your encoder produces a regular cadence I-frames. Take a look at the multiple samples in the repository and look through the docs.

MushMal
  • 109
  • 2
  • Thank you for the reply. Sorry I missed an important fact in my original question - all these hundreds of 0-length fragments have the same timestamp (that's why I'm not too worried about them). Seems either our upstream component or gstreamer or the KVS plugin is trying repeatedly to send the same fragment. Does the new info change your thinking on this? – Tony Jiang Sep 02 '20 at 18:26