Nvvideoconvert segfault after several hours

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, , ,
1

• Hardware Platform : Jetson Orin embedded in Connect Tech Anvil
• DeepStream Version : 6.3.0
• JetPack Version : 5.1.2
• TensorRT Version : 8.5.2

Hello,

I designed a c++ gstreamer pipeline to process data from 6 camera (30 fps, 1920*1280) on Jetson orin.
This pipeline aimed to do the inference on 6 cameras and encode the stream with h264 encoding.
After creating and starting the pipeline, it crashes about several hours. It seems to always crash on the gst-element nvvideoconvert (convert UYVY to RGBA before inference) at the start of the pipeline.

Snippet of the code where the nvvideconvert element is created :

    GstElement* cam_src = m_pipeline->createAndAddElement("v4l2src");
    GstElement* format_convert = m_pipeline->createAndAddElement("nvvideoconvert");
    GstElement* identity = m_pipeline->createAndAddElement("identity");

    auto* l_src_pad = gst_element_get_static_pad(cam_src, "src");
    gst_pad_add_probe(l_src_pad,
                      GST_PAD_PROBE_TYPE_BUFFER,
                      receiveCameraDataProbe,
                      m_camera_data[cam_index].get(),
                      NULL);

    string device = "/dev/video" + to_string(cam_index);
    g_object_set(cam_src, "device", device.c_str(), NULL);


    GstCaps* caps = gst_caps_new_simple("video/x-raw",
                                        "format",
                                        G_TYPE_STRING,
                                        "UYVY",
                                        "width",
                                        G_TYPE_INT,
                                        m_config_handler->m_config_sensors.m_width,
                                        "height",
                                        G_TYPE_INT,
                                        m_config_handler->m_config_sensors.m_height,
                                        "framerate",
                                        GST_TYPE_FRACTION,
                                        m_config_handler->m_config_sensors.m_framerate,
                                        1,
                                        NULL);
    if (!m_pipeline->elementLinkFiltered(
          cam_src,
          format_convert,
          caps,
          "RUNTIME_SOURCE_CREATION v4l2src->nvvideoconvert for camera" + to_string(cam_index)))
    {
        return false;
    }

    if (!m_pipeline->elementLinkFiltered(
          format_convert,
          identity,
          "video/x-raw(memory:NVMM), format=(string)RGBA",
          "RUNTIME_SOURCE_CREATION nvvideoconvert->identity for camera" + to_string(cam_index)))
    {
        return false;
    }

Here is a screenshot of the coredump, showing that the segfault occurs inside the nvvideoconvert element :

image
image1169×614 138 KB

When I run the pipeline with 3 cameras, the pipeline run without any crashes. May it be a performance issue ? Is there different set of parameters of the nvvideoconvert that I can try to avoid the crash from occuring ?

I can provide more information if necessary.

Kind regards,

Grégoire

3
  1. You can try to refer to our FAQ to check the memory status.

2.Could you try to run your pipeline with gst-launch-1.0 command first to check if there is crash?