Speeding up a queue

[simo zz]

Hello, 
As promised, I followed the Nicolas's suggestion and now it's working fine.Regards.


 


    El Sábado 9 de septiembre de 2017 16:25, "simon.zz at yahoo.com" <simon.zz at yahoo.com> escribió:
 

 Hello Martín,
Thank you for tour hint. I will keep in mind that.

At the moment, since I want to write the content of the map.data to a pipe, I tested the fdsink and the filesink instead of the appsink since they write data without the need to do it manually.

The fdsink give me problems.. It actually overload the program and the video is not written any more after the first chunk of data.
filesink works far better.
The only problem I see with using filesink with a pipe, is that if the reader closes the stream, gstreamer stops/crashes.

However I don't discard to use the appsink again after your hints. I will test it and report the results.

Regards,
Simon


--------------------------------------
El vie, 8/9/17, Nicolas Dufresne <nicolas at ndufresne.ca> escribió:

 Asunto: Re: Speeding up a queue
 Para: "Discussion of the development of and with GStreamer" <gstreamer-devel at lists.freedesktop.org>, "simo zz" <simon.zz at yahoo.com>
 Fecha: viernes, 8 de septiembre, 2017 19:21
 
 Le vendredi 08 septembre 2017 à
 09:11 +0000, Martin Vachovski a écrit :
 > Hi Simon,
 > 
 > Three suggestions in the order of
 relevance:
 > 
 > 1) I
 think there is no need to explicitly call 
 > gst_buffer_ref(buffer)-- the
 gst_sample_get_buffer should do it
 >
 internally
 
 No it
 "transfer none".
 
 > 
 > 2) Are you sure that
 the performance is the issue here?
 > Try
 changing the condition whether you access the data
 > and see when it stops working, like:
 > if(!(buffCntr % 20)) // every 20th
 > then 
 > if(!(buffCntr %
 15)) // every 15th
 > and see if for
 sufficiently small denominator it stops working
 > 
 > 3) Also, try calling
 the "unref" methods in the opposite order as
 the
 > "get/ref" methods: 
 > You have 
 >
 ref(sample)
 > ref(buffer)
 > But you release the sample first
 > unref(sample)
 >
 unref(buffer)
 > I think it should be:
 > unref(buffer)
 >
 unref(sample)
 
 This is
 usually not required with ref-counting in general. Since
 if
 there dependent object will have a ref on
 the other, which guaranty the
 free order for
 you.
 
 > 
 >
 Hope that helps
 > Martin
 > 
 > From:
 gstreamer-devel <gstreamer-devel-bounces at lists.freedesktop.org>
 > on behalf of simo zz <simon.zz at yahoo.com>
 > Sent: Thursday, September 7, 2017 5:00
 PM
 > To: Discussion of the Development of
 and With GStreamer
 > Subject: Speeding up
 a queue
 >  
 > Hello,
 
 > I need to record a video from the USB
 camera and process the raw-data 
 > at the
 same time, so I successfully implemented this pipeline in
 C:
 > 
 > v4l2src
 device=/dev/webcam ! videoconvert ! video/x-
 > raw,width=544,height=288,framerate=10/1 !
 tee name=t ! queue
 > ! v4l2h264enc !
 h264parse ! mp4mux ! filesink location=video.mp4 t. !
 > queue ! appsink
 > 
 > I can access to raw-data from USB camera
 is read using the "new-
 >
 sample" signal of appsink in a function called
 "dataProbe", code as
 >
 follows:
 > 
 >
 GstFlowReturn dataProbe(GstElement *source)
 > {  
 >     //
 buffCntr is just a counter to choose when I want to
 process
 > the raw-data
 >     if (!(buffCntr % 10))
 >     {   
 >     
    GstMapInfo map;
 >        
 GstSample *sample =
 >
 gst_app_sink_pull_sample(GST_APP_SINK(source));
 >         GstBuffer *buffer =
 gst_sample_get_buffer(sample);
 >     
    gst_buffer_ref(buffer);
 >        
 gst_buffer_map (buffer, &map, GST_MAP_READ);
 >         g_print("Map size:
 %d\n", map.size);
 > 
 >         /*int8_t fd =
 open("/path/to/someFile", O_CREAT | O_WRONLY);
 >         if (fd != -1)
 >         {
 >     
        if(write(fd, map.data, map.size) == map.size)
 >                
 g_print("Written whole data\n");
 >             else
 >                
 g_print("Cannot write whole data\n");
 >         }
 >     
    close(fd);*/
 >        
 gst_buffer_unmap(buffer, &map);
 > 
        gst_sample_unref(sample);
 > 
        gst_buffer_unref(buffer);
 > 
 >         //buffCntr = 0;
 >     }
 > }
 > 
 > But the execution of
 the function dataProbe is so critical that if I
 > set buffCntr to zero (as shown in the last
 commented instruction of
 > the function)
 even without writing map.data to some file, the whole
 > program hangs, so the mp4 video is not
 written anymore.. 
 > But I need both
 raw-data and mp4 video.
 > 
 > What can I do to 'speed up' this
 queue and let my gstreamer program
 > to
 record the video and write the raw data somewhere ? Pthread
 ?
 > There is some helpful feature already
 available from gstreamer
 > library to
 handle this situation ?
 > 
 > The device which run the application is an
 embedded board with a quad
 > core CPU.
 > 
 > Thank you in
 advance.
 > Regards,
 >
 Simon
 >
 _______________________________________________
 > gstreamer-devel mailing list
 > gstreamer-devel at lists.freedesktop.org
 > https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel_______________________________________________
 gstreamer-devel mailing list
 gstreamer-devel at lists.freedesktop.org
 https://lists.freedesktop.org/mailman/listinfo/gstreamer-devel
 
 -----Adjunto en línea a continuación-----
 
 

   
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/gstreamer-devel/attachments/20170912/dd2bd028/attachment.html>