[
https://issues.apache.org/jira/browse/MESOS-9595?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16947125#comment-16947125
]
Patrick commented on MESOS-9595:
--------------------------------
This issue means that at runtime we end up with a path on the container like
this
{{/run/mesos/isolators/gpu/nvidia_418.87.00/}}
>From what I can see, there is no way for me to configure this to something
>generic like
{{/run/mesos/isolators/gpu/nvidia/}}
so that I can use softlinks to find the nvidia libraries on the container.
My goal is to be able to change the undelying library versions without having
to change the marathon and so mesos application configuration. Short of writing
a custom isolator.
Is there a better solution I am missing?
> Use `libnvidia-container` in the `gpu/nvidia` isolator.
> -------------------------------------------------------
>
> Key: MESOS-9595
> URL: https://issues.apache.org/jira/browse/MESOS-9595
> Project: Mesos
> Issue Type: Improvement
> Components: containerization
> Reporter: Chun-Hung Hsiao
> Priority: Major
>
> Nvidia uses
> [nvidia-container-runtime|https://github.com/NVIDIA/nvidia-container-runtime]
> in their new [nvidia-docker 2.0|https://github.com/NVIDIA/nvidia-docker]
> images, which consists of runc hooks that uses
> [libnvidia-container|https://github.com/NVIDIA/libnvidia-container] to
> prepare the container runtime environment, including the binaries and
> libraries. We should update our {{gpu/nvidia}} isolator to use
> {{libnvidia-container}} (either through its CLI or using the library API
> directly) to prepare the container runtime environment instead of hard-coding
> the logic from nvidia-docker 1.0 in Mesos.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
