| Summary: | nvidia-docker version 2 support | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Ladislav Zitka <archenroot> |
| Component: | Current packages | Assignee: | Gentoo Linux bug wranglers <bug-wranglers> |
| Status: | RESOLVED NEEDINFO | ||
| Severity: | normal | CC: | jstein |
| Priority: | Normal | ||
| Version: | unspecified | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
|
Description
Ladislav Zitka
2019-05-31 21:01:46 UTC
which packages exactly? please write the full category/package name in the summary Ah, sry, sure. I open this request to provide in Gentoo support for nvidia docker project which allows one to use GPU devices within the docker containers. There exist 2 ebuilds for version 1, which is becoming now obsolete: 1. https://data.gpo.zugaina.org/cynede/app-admin/nvidia-docker/ 2. another binary version is in ros-overlay under name:sci-libs/nvidia-docker-bin But this request is to support in Gentoo version 2. Now the problem is the packaging supports at moment only deb and rpm based distributions. Main site is here: https://nvidia.github.io/nvidia-docker/ Wiki: https://github.com/NVIDIA/nvidia-docker/wiki there are discussions about how to build for other distros: https://github.com/NVIDIA/nvidia-docker/issues/553 https://github.com/NVIDIA/nvidia-container-runtime/issues/9 Especially this part from second issue above is important, but I didn't test it yet: They are not private repos. They are public package repositories from NVIDIA. They are just not controlled by Red Hat/Canonical. For libnvidia-container, use the tarball here: https://github.com/NVIDIA/libnvidia-container/releases/tag/v1.0.0-rc.2 And instructions are here: https://nvidia.github.io/libnvidia-container/ For nvidia-container-runtime-hook, there is a static binary here: https://github.com/NVIDIA/nvidia-container-runtime/releases/tag/v1.4.0-1 Install it somewhere in your PATH. If you are using a container tool that can handle OCI hooks directly like runc, CRI-O or podman, you can stop here. You are ready to go. If you are using docker, there is one extra step: install runc after applying this patch: https://github.com/NVIDIA/nvidia-container-runtime/blob/master/runtime/runc/3f2f8b84a77f73d38244dd690525642a72156c64/0001-Add-prestart-hook-nvidia-container-runtime-hook-to-t.patch Rename the generated binary from runc to nvidia-container-runtime then register it to the docker daemon: https://github.com/nvidia/nvidia-container-runtime#docker-engine-setup |
