Skip to content
This repository has been archived by the owner on Jan 22, 2024. It is now read-only.

Invalid cross-device link: how to reinstall NVIDIA drivers in a specific location? #335

Closed
aTnT opened this issue Mar 9, 2017 · 3 comments

Comments

@aTnT
Copy link

aTnT commented Mar 9, 2017

I am having problems similar to issue #133 which in my case and as mentioned here are due to the need to have both NVIDIA drivers and the Docker plugin volumes in the same partition (this is not my case).

How can i remove and reinstall the NVIDIA drivers in a specific location? I normally use the command below to install the drivers:

sudo apt-get install nvidia-xxx

This is my output of journalctl -n -u nvidia-docker :


mrt 09 11:47:45 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 11:47:45 Serving remote API at localhost:3476
mrt 09 11:50:33 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 11:50:33 Received activate request
mrt 09 11:50:33 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 11:50:33 Plugins activated [VolumeDriver]
mrt 09 11:50:33 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 11:50:33 Received create request for volume 'nvidia_driver_367.57'
mrt 09 11:50:33 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 11:50:33 Error: link /usr/lib/nvidia-367/bin/nvidia-cuda-mps-control /data/nvidia-docker/nvidia_driver/367.57/bin/nvidia-cuda-mps-control: invalid cross-device link
mrt 09 12:03:53 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 12:03:53 Received activate request
mrt 09 12:03:53 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 12:03:53 Plugins activated [VolumeDriver]
mrt 09 12:03:53 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 12:03:53 Received create request for volume 'nvidia_driver_367.57'
mrt 09 12:03:53 andre-thinkstation-P710 nvidia-docker-plugin[1460]: /usr/bin/nvidia-docker-plugin | 2017/03/09 12:03:53 Error: link /usr/lib/nvidia-367/bin/nvidia-cuda-mps-control /data/nvidia-docker/nvidia_driver/367.57/bin/nvidia-cuda-mps-control: invalid cross-device link

As seen above /data/nvidia-docker is the location of my nvidia-docker volume, and the contents of systemctl edit nvidia-docker are:


[Service]
ExecStart=
ExecStart=/usr/bin/nvidia-docker-plugin -s $SOCK_DIR -d /data/nvidia-docker

@flx42
Copy link
Member

flx42 commented Mar 9, 2017

Instead of trying to install the driver to a different location (not a great idea since it's installed through a deb package), you should instead modify the location of the nvidia-docker volumes.

@aTnT
Copy link
Author

aTnT commented Mar 9, 2017

The point is that i am out of space and i need to set the volumes and containers in an external partition (and physical drive) with sufficient space.

@flx42
Copy link
Member

flx42 commented Mar 9, 2017

In this case, remove the current NVIDIA driver (purge all the packages). Then install the driver through our .run installers from this page: http://www.nvidia.com/object/unix.html
If you look at the advanced options, you will be able to set the install path.

@3XX0 3XX0 closed this as completed Jun 13, 2017
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants