Proper way to stop nvarguscamerasrc and release all resources?

Robotics & Edge Computing Jetson Systems Jetson Orin NX
1

Hi NVIDIA team,

I’m looking for the recommended way to gracefully stop an nvarguscamerasrc pipeline and ensure that all Argus resources are released.

My pipeline is simply:



gst-launch-1.0 nvarguscamerasrc ! \
'video/x-raw(memory:NVMM), width=4000, height=3000, format=NV12, framerate=30/1' ! \
fakesink

In my application, I try to stop the pipeline by sending an EOS event and waiting for the EOS message from the bus before setting the pipeline to GST_STATE_NULL.



void gst_message_handler(GstElement* pipeline, int gst_msg) {
    GstBusPtr bus(gst_element_get_bus(pipeline));

    GstMessagePtr msg(
        gst_bus_timed_pop_filtered(
            bus.get(),
            GST_CLOCK_TIME_NONE,
            static_cast<GstMessageType>(gst_msg)));

    if (msg != nullptr) {
        if (GST_MESSAGE_TYPE(msg.get()) == GST_MESSAGE_EOS) {
            logging("Receive EOS");
            gst_element_set_state(pipeline, GST_STATE_NULL);
        } else if (GST_MESSAGE_TYPE(msg.get()) == GST_MESSAGE_ERROR) {
            GError* err = nullptr;
            gchar* debug_info = nullptr;

            gst_message_parse_error(msg.get(), &err, &debug_info);

            logging("Gst Message Error: {}", err->message);

            g_clear_error(&err);
            g_free(debug_info);
        }
    }
}

auto StopGstThread = std::thread(
    [pipeline = _main_pipeline.get()]() {
        gst_message_handler(pipeline, GST_MESSAGE_EOS);
    });

gst_element_send_event(_main_pipeline.get(), gst_event_new_eos());

if (StopGstThread.joinable()) {
    StopGstThread.join();
}

g_main_loop_quit(_main_loop.get());

However, the EOS message never arrives. The thread blocks forever in:



gst_bus_timed_pop_filtered(..., GST_CLOCK_TIME_NONE, GST_MESSAGE_EOS);

As a result, the pipeline never transitions to GST_STATE_NULL.

My questions are:

  1. Is sending an EOS event the correct way to stop a pipeline using nvarguscamerasrc?
  2. Does nvarguscamerasrc support propagating EOS, or is there a different recommended shutdown sequence?
  3. What is the proper procedure to ensure all camera/Argus resources are released before restarting the pipeline?

Any guidance or example code would be greatly appreciated. Thanks!

3

hello triconguyen251,

usually, we’ll have -e included in the gst pipeline. with the -e options, it will send an EoS at shutdown.

besides..
you may refer to Argus samples, such as.. Argus/public/samples/userAutoExposure.
in general, here’re the flow of app shutdown process.

     └── Shutdown after requested frame count                                                                                                                     
          ├─ Destroy Request
          ├─ Destroy OutputStreamSettings
          ├─ Destroy EventQueue
          ├─ Destroy CaptureSession                                                                                                                               
          │    └─ Repeating capture stops implicitly
          ├─ Destroy EGL OutputStream                                                                                                                             
          ├─ Stop and join PreviewConsumerThread
          │    ├─ Delete GL program/textures                                                                                                                      
          │    └─ Destroy GL context
          ├─ Destroy CameraProvider  
          ├─ Shut down window                                                  
          └─ Clean up EGL display
4

Hi @JerryChange ,

I am running pipeline with my Cpp programing with

gst_parse_launch();

and i send EOS then listen when the pipeline catch the EOS event, but it has been never called

void stop_gstreamer() {

gst_element_send_event(\_main_pipeline.get(), gst_event_new_eos());

}

I have add listen to the bus before i start the pipeline, here is my listener

gboolean GstreamerInterface::bus_callback(GstBus *bus, GstMessage *msg) {

switch (GST_MESSAGE_TYPE(*msg*)) {

case GST_MESSAGE_EOS:

logging_DEBUG(“EOS received on bus, quitting main loop.\n”);

if (_main_pipeline) {

g_main_loop_quit(_main_loop.get());

gst_element_set_state(GST_ELEMENT(_main_pipeline.get()), GST_STATE_NULL);

            // main_pipeline is a smart pointer

            // already hasGstElementDeleter => comment unref manual code

            // gst_object_unref(GST_OBJECT(\_main_pipeline.get()));

g_source_remove(bus_watch_id);

_main_pipeline.reset();

_is_pipeline_running.store(false, std::memory_order_relaxed);

logging_INFO(“Main pipeline stopped”);

        }

break;

case GST_MESSAGE_ERROR: {

        GError \*err   = nullptr;

        gchar  \*debug = nullptr;

gst_message_parse_error(msg, &err, &debug);

logging_ERROR(“Error: %s\n”, err->message);

g_error_free(err);

g_free(debug);

        // g_main_loop_quit(\_main_loop.get());

break;

    }

default:

break;

}

return TRUE; // keep watching

}

How to send the EOS to the pipeline and wait until it terminate correctly and completely clear all resource ?

5

hello triconguyen251,

-e option is a gst-launch-1.0 command-line option; it must not be placed in a gst_parse_launch() string. cpp code should be.. gst_element_send_event(pipeline, gst_event_new_eos());
please see-also similar discussion thread, Topic 176364 for reference.