Dockerfile in building triton

• Problem Hello I am using triton inference server. And when i using docker image 21.07. I have to install a package to fix issues about related to libGL.so.1. This is link issues: https://github.com/conda-forge/pygridgen-feedstock/issues/10#issuecomment-365914605 I rebuild a triton image with docker. Refer to : https://github.com/triton-inference-server/server/blob/r21.07/docs/build.md#building-with-docker
• Solution Before build triton with ./build.py command. I add a line into build.py file at line 706 in a create dockerfile function

RUN apt-get -yqq update && apt -yqq install libgl1-mesa-glx

I think this is a common problem. So, should we add the line above into build.py file of the repo.

[csvance]

@gioipv If you install opencv through pip, you can install opencv-python-headless in order to avoid the dependency. conda-pack works just fine with pip packages as well.

Conda OpenCV also uses OpenMP for multi threading which doesn’t play nicely with many computing paradigms without a ton of extra configuration and debugging. The pip packaging of OpenCV uses vanilla pthreads, which are much easier to work with and play nicer with other things.

[Tabrizian]

It looks like it is more a problem with the way conda packages the dependencies. We can’t change Triton container to include dependencies for Python packages. I think the only work around for now would be to create another Docker container that contains the dependencies required for your Python environment.