Changing config file path property in runtime for nvinfer, nvdsanalytics elements

[apisaruthan]

• Hardware Platform (Jetson / GPU) - Tesla T4 GPU
• DeepStream Version - 7.0
• TensorRT Version - 8.6.1.6
• NVIDIA GPU Driver Version (valid for GPU only) - 535.171.04
• CUDA - 12.2
• Issue Type( questions, new requirements, bugs) - Question

I had a requirement where I need to set the model config file path property in runtime for nvinfer elment continously. For every video I am processing, I will be having model config in a different file path, SO I need to set that each time before processing a video at runtime.

So I modified the code and It is working pretty good as expected, But I am having this warning sometimes

deepstream-2     | 0:09:44.615318721     1 0x791ceb0520b0 WARN                 nvinfer gstnvinfer_impl.cpp:346:notifyLoadModelStatus:<primary-nvinference-engine> warning: [UID 1]: Load new model:/opt/nvidia/deepstream/deepstream-7.0/app/fast_api_server/tmp/05503eff/configs/model_config.txt failed, reason: Trying to update model too frequently

So I want to know whether, is it okay to have this warning?, Or is it not good to set these config file path properties for nvinfer element/others in the runtime?. Do we have any breaking issues if we run it for a long time let’s say if we have to process 1000 videos?

Would be nice to have a answer that will help to decide whether to use this approach(since this is working) or not? Thanks in advance

[fanzh]

the log means “Check if a model is loaded but not yet being used for inferencing.”. seems that the model was updated frequently. nvinfer plugin and low-level lib are opensource. please refer to DsNvInferImpl::loadModel.
nvinfer supports OTA functionality. Test5 app can update the models in the running pipeline on-the-fly.

[apisaruthan]

Thanks for the information. And here I am actually trying to load the same model. But the model path is specified in the model config file and I am changing the model config path on runtime. Is this warning harmful for that particular scenario?

[fanzh]

As the code DsNvInferImpl::loadModel shown, this log is an abnormal case. The model will not be loaded because the function will return in advance.

[apisaruthan]

Thanks. And where can I find the code for nvinfer DsNvInferImpl::loadModel ?

[fanzh]

\opt\nvidia\deepstream\deepstream\sources\gst-plugins\gst-nvinfer\gstnvinfer_impl.cpp. the function includes the following code.

    notifyLoadModelStatus (ModelStatus {
        NVDSINFER_UNKNOWN_ERROR, modelPath, "Trying to update model too frequently"}

[apisaruthan]

Thanks

[fanzh]

Sorry for the late reply, Is this still an DeepStream issue to support? Thanks!