Maybe it is not a memory leak issue because the time varies from 6 minutes to 14+ hours. Did you see any error message in the dmesg output? Thanks Haihao > Is what I suspect but I have also a decoder running in the same process and I > never saw this failure in other parts of gstreamer-vaapi. > > Could you provide any advice on how to debug this? I fear that run inside > Valgrind will be not feasible. > > Thanks > > > > > On Thu, Jul 26, 2018 at 10:22 AM Víctor Jáquez <vjaquez(a)igalia.com&gt; wrote: > > > On Thu, 26 Jul 2018 at 07:50, Matteo Valdina wrote: > > > > Hi, > > > > I'm facing a bug with the intel-vaapi-driver and gstreamer-vaapi. > > > > > > > > My pipeline is something like: > > > > > > > > v4l2src -> vaapipostproc-> vaapih264enc -> rtp > > > > > > > > This is a 4K@30 stream in NV12 and encoded in High profile at 6Mbps. > > > > The VPP is there when there is a need to rescale the frame. > > > > > > > > After an undefined enough of time that it could be from 6 minutes to 14 > > > > hours (or more), vaapi fails to create a SurfaceID and the pipeline > > > stopped. > > > > > > > > If the VPP is processing data (unfortunately I noticed that if you set > > > the > > > > rescale mode to fast it pass through the rescaler without needs to > > > rescale > > > > the frame) it fails with these error: > > > > > > > > Jul 16 22:01:46 localhost gstd[2669]: failed to apply VPP filters (error > > > 2) > > > > Jul 16 22:01:46 localhost gstd[2669]: error: Internal data stream error. > > > > Jul 16 22:01:46 localhost gstd[2669]: error: streaming stopped, reason > > > > error (-5) > > > > > > > > I added additional logs in this failure is reported in the vaEndPicture > > > > with invalid SurfaceID. > > > > Adding more logs I noticed that gstreamer-vaapi fails to create the > > > surface > > > > and pass to the rescaler a surfaceid -1. > > > > > > > > If I disable the VPP the issue is not resolved but moved to the > > > > vaapih264enc with the following error: > > > > > > > > Jul 23 17:22:03 localhost gstd[2329]: create H264 picture failed, > > > > frame timestamp:3:03:07.581673496 > > > > Jul 23 17:22:03 localhost gstd[2329]: failed to process reordered frames > > > > Jul 23 17:22:03 localhost gstd[2329]: failed to encode frame 15901 > > > (status -2) > > > > Jul 23 17:22:03 localhost gstd[2329]: error: Internal data stream error. > > > > Jul 23 17:22:03 localhost gstd[2329]: error: streaming stopped, reason > > > > error (-5) > > > > > > It this case it look you are in a memory exhaustion condition, where the > > > encoder > > > cannot allocate further memory. A memory leak somewhere? > > > > > > In the case of the VPP looks similar, as if the driver couldn't allocate a > > > new > > > surface. > > > > > > > > > > > Also here it looks like the vaapi fails to create a valid SurfaceID. > > > > > > > > I'm currently on gstreamer 1.14.0, libva 2.1.0, intel-vaapi-driver > > > 2.1.0, > > > > kernel 4.12.14 and the CPU is a Intel(R) Celeron(R) CPU 3965U @ 2.20GHz. > > > > > > > > Any suggestion on how to debug this issue? > > > > Enable LIBVA_TRACE is not viable because it will exhaust the space in a > > > > couple of hours. > > > > I should try to update to the latest libva, intel-vaapi-driver? > > > > > > > > Best > > > > Matteo Valdina > > > > > > > _______________________________________________ > > > > intel-vaapi-media mailing list > > > > intel-vaapi-media(a)lists.01.org > > > > https://lists.01.org/mailman/listinfo/intel-vaapi-media > > > > > > > > > -- > > “There are two ways of constructing a software design: One way is to make it > > so simple that there are obviously no deficiencies, and the other way is to > > make it so complicated that there are no obvious deficiencies. The first > > method is far more difficult.” > > - Tony Hoare > > > _______________________________________________ > intel-vaapi-media mailing list > intel-vaapi-media(a)lists.01.org > https://lists.01.org/mailman/listinfo/intel-vaapi-media