Unable to create egl x11 window in new nvidia docker container (Was: nvdec: "OpenGL context is not CUDA-compatible, fallback to system memory" after creating glimagesink in 1.20.3)

[Michiel Konstapel]

I've worked around it by using xvimagesink instead of glimagesink, but 
I've been trying to narrow this down. It does not appear to be the 
gstreamer version but the environment.

My previous message was mistaken: it used to be Ubuntu 18.04 and the new 
version is 20.04. I'm not sure I'm on the right track but this seems 
relevant: I am building and running gstreamer inside Docker containers 
based on the nvidia/cuda images. When I build/run inside 
nvidia/cuda:10.2-cudnn7-devel (Ubuntu 18.04) then

     gst-launch-1.0 videotestsrc ! glimagesink

says

     Got context from element 'sink': gst.gl.GLDisplay=context, 
gst.gl.GLDisplay=(GstGLDisplay)"\(GstGLDisplayX11\)\ gldisplayx11-0";

And it shows a window.

However, on the nvidia/cuda:11.6.2-cudnn8-devel-ubuntu20.04 base image, 
it says

     Got context from element 'sink': gst.gl.GLDisplay=context, 
gst.gl.GLDisplay=(GstGLDisplay)"\(GstGLDisplayEGL\)\ gldisplayegl0";

and it runs, but doesn't show a window.

However, xvimagesink works (shows a window) and f.e. glmark2 from inside 
the container also shows a window - so X11 appears to be available.

I've tried to force it:

     GST_GL_PLATFORM=egl GST_GL_WINDOW=x11 gst-launch-1.0 -v 
videotestsrc ! glimagesink

But that returns:

   (gst-launch-1.0:29): GStreamer-GL-CRITICAL **: 11:16:45.500: 
gst_gl_display_egl_from_gl_display: assertion 'native_display != 0' failed
ERROR: from element 
/GstPipeline:pipeline0/GstGLImageSinkBin:glimagesinkbin0/GstGLImageSink:sink: 
Failed to create EGLDisplay from native display
Additional debug info:
../subprojects/gst-plugins-base/ext/gl/gstglimagesink.c(1101): 
_ensure_gl_setup (): 
/GstPipeline:pipeline0/GstGLImageSinkBin:glimagesinkbin0/GstGLImageSink:sink

Am I missing some kind of dependency that will let gstreamer create a GL 
X11 window?

I am setting the DISPLAY environment variable to the same as the host's 
and binding .Xauthority into the container.

Kind regards,
Michiel

On 18-07-2022 10:38, Michiel Konstapel wrote:
> After updating from gstreamer from 1.18.3 to 1.20.3 (and Ubuntu from 
> 20.04 to 22.04) my pipeline no longer works:
>
> nvdec gstnvdec.c:1770:gst_nvdec_check_cuda_device_from_context: CUDA 
> call failed: CUDA_ERROR_INVALID_GRAPHICS_CONTEXT, invalid OpenGL or 
> DirectX context
> nvdec gstnvdec.c:1822:gst_nvdec_ensure_gl_context:<nvh264dec0> Current 
> OpenGL context is not CUDA-compatible
> nvdec gstnvdec.c:646:gst_nvdec_negotiate:<nvh264dec0> OpenGL context 
> is not CUDA-compatible, fallback to system memory
>
> (Then negotiation fails because downstream expects memory:GLMemory)
>
> However, this only happens when there is a glimagesink in the 
> pipeline. If I take that out, then apparently nvh264dec gets a context 
> it likes and all is well.
>
> This gst-launch pipeline works, and prints helpful messages indicating 
> both get different, suitable contexts?
>
> gst-launch-1.0 v4l2src device=/dev/video0 ! h264parse ! nvh264dec ! 
> "video/x-raw(memory:GLMemory)" ! glimagesink
> Setting pipeline to PAUSED ...
> error: XDG_RUNTIME_DIR not set in the environment.
> Pipeline is live and does not need PREROLL ...
> Got context from element 'sink': gst.gl.GLDisplay=context, 
> gst.gl.GLDisplay=(GstGLDisplay)"\(GstGLDisplayEGL\)\ gldisplayegl0";
> Got context from element 'nvh264dec0': gst.cuda.context=context, 
> gst.cuda.context=(GstCudaContext)"\(GstCudaContext\)\ cudacontext0", 
> cuda-device-id=(int)0;
> Pipeline is PREROLLED ...
> Setting pipeline to PLAYING ...
>
> Any clue on how to fix this, or where to start debugging?
>
> Thanks in advance,
> Michiel