Bug related to egl stream and nvEnc Apis With CUDA toolkit 11.7

I encountered an NVIDIA/CUDA driver bug issue.

My System Info:
intel nuc 11 machine(NUC11PHKi7C,幻影峡谷),
linuxs OS, ubuntu 20.04,
NVIDIA graphics card,
Graphics Processor: NVIDIA GeForce RTX 2060
CUDA Cores: 1920
Total GPU Memory: 6144MB

My Application includes several processes, playing the role of producer or consumer of egl stream.
My Application involved several kinds of apis, such as opengl, opengles, cuda, nvenc APIs.
In egl stream producer process, use opengl renderer with egl stream surface (eglCreateStreamProducerSurfaceKHR).
In egl stream consumer process, connect and consume the egl stream.
The APis used in consumer process includes and not limited to:

At the egl stream producer process, when stream-related surface is destroyed, eglDestroyStreamKHR is invoked to destroy egl stream.
At the same time, at the egl stream consumer process,
need to invoke cuEGLStreamConsumerDisconnect and eglDestroyStreamKHR
to destroy stream to avoid memory leak.

When I use CUDA toolkit 11.7(nvidia/cuda driver 515.43.04/11.7), or CUDA toolkit 11.6(nvidia/cuda driver 510.39.01/11.6),
then, once ‘destroy egl stream’ occured, the egl stream consumer process/thread will block in nvEncEncodePicture or nvEncUnregisterResource api invoke,
and can’t come out. At the same time, there seems no error reported.

When I use CUDA toolkit 11.8(nvidia/cuda driver 520.61.05/11.8), every thing work well and no problem.

So I guess is there a serious bug in the CUDA toolkit 11.7(nvidia/cuda driver 515.43.04/11.7) ?
This is also my purpose for this contact with NVIDIA support or driver developer.
I just want to confirm is there a serious bug in the CUDA toolkit 11.7, according to my above problem description.
Thanks a lot.