Please provide complete information as applicable to your setup.
**• Hardware Platform GPU - 2 x A2
**• DeepStream Version 6.3
**• TensorRT Version 8.5.3.1
**• NVIDIA GPU Driver Version - 535.129.03
**• Issue Type - bug
I’m trying to debug an issue that I am struggling to find the route cause of.
I have a deepstream 6.3 pipeline (created with Python bindings) that roughly goes:
nvurisrcbin -> muxer -> pgie -> tracker -> analytics -> nvstreamdemux -> nvvideoconvert -> nvdsosd -> nvvideoconvert -> h264parse -> rtph264pay -> udpsink
I am then using GstRtspServer.RTSPServer to serve RTSP streams. My pipeline takes 7 input RTSP streams from live cameras on my local network and serves 7 annotated RTSP streams via a single GstRtspServer.
This works fine for a period of time (around 30 mins), after which the pipeline seems to falter, the streams die and the console output shows periodic “Resetting source” messages from the rtsp-reconnect-interval configured in nvurisrcbin.
The load on the GPU is a steady 60% or so, which drops to around 40% after it gets into this state.
Can anyone think of a good way to debug this? I have a feeling that it is something to do with time synchronisation between the streams in my pipeline. Are there elements or configuration changes that I can make to try to improve the robustness? I will also try upgrading to 6.4 I think.