Update CUDA repository GPG keys on Ubuntu 18.04

[patrick-5546]

Current the following command does not work: ue4-docker build 4.19.2 --cuda=10.0 --no-full. This is because the CUDA repository GPG keys on Ubuntu 18.04 have been rotated. I have found a workaround for a dockerfile that uses the same base image, but am unsure which dockerfiles need to be updated in this repository; would I just have to add those 2 lines after line 2 of https://github.com/adamrehn/ue4-docker/blob/master/ue4docker/dockerfiles/ue4-build-prerequisites/linux/Dockerfile?

[TBBle]

That workaround appears to just disable the CUDA repositories. If we don't need anything from them, that's fine (although then why are we using an nvidia base image at all?).

Otherwise, we might want to instead use this process which updates the signing keys instead.

Whichever way we go, I'd put the workaround immediately before the first RUN apt-get update step in the ue4-build-prerequisites Dockerfile so it's clear why it's there, and what it relates to.

[patrick-5546]

I have confirmed that the following works on cudagl images built on ubuntu 18.04 and 20.04:

RUN [ ! -e /etc/apt/sources.list.d/cuda.list ] \
    || rm /etc/apt/sources.list.d/cuda.list
RUN [ ! -e /etc/apt/sources.list.d/nvidia-ml.list ] \
    || rm /etc/apt/sources.list.d/nvidia-ml.list
RUN apt-key del 7fa2af80 \
    && apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/3bf863cc.pub

cuda.list and nvidia-ml.list doesn't exist on 20.04 and beyond.

[TBBle]

If you're using the last RUN line, then you won't need the previous two, the point of that last line is to make the first two unnecessary, without losing access to those repos in the meantime.

[patrick-5546]

Without the first 2 RUN lines, apt-get update still errors out:

W: GPG error: https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY F60F4B3D7FA2AF80
E: The repository 'https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64  Release' is not signed.

[TBBle]

Ah yes, the machine-learning repo is abandoned, so we do still need to delete /etc/apt/sources.list.d/nvidia-ml.list, as it has not been resigned with the new key.

[TBBle]

I'd also suggest that after enrolling the new key and apt get update being done, doing an apt-get install cuda-keyring (with all the ceremony of the other apt-get install calls) to bring the package state into line with the actual state. (That might need to be apt upgrade -y cuda-keyring if apt-get install doesn't auto-upgrade the named package.

[patrick-5546]

I just found out that nvidia:cuda replaces nvidia:cudagl and its images use the updated cuda repo signing key: https://hub.docker.com/r/nvidia/cuda.

Searching through the repository for cudagl, the following files should be updated to use the new image:

• docs/advanced-build-options.adoc
• docs/available-container-images.adoc
• ue4docker/infrastructure/BuildConfiguration.py