"Error: unsupported Docker version"

[stefanseefeld]

Simply trying to follow the documented steps to install and test nvidia-docker yields the above error message. I'm running docker 1.11.0 (build 4dc5990), on an Ubuntu 14.04 system.

[flx42]

Docker 1.11.0 was released yesterday, 30 minutes later I added support to nvidia-docker: 839fe26

It's not part of the deb package or the tarball yet, we will create a new release soon.
In the mean time you can build the binaries, the tarball or the deb manually:
https://github.com/NVIDIA/nvidia-docker/wiki/Installation#building-from-sources

[stefanseefeld]

Oh, sounds like bad timing ! (I hadn't even realized that I upgraded my docker installation this morning to such a fresh version. :-) )
Thanks for addressing this so quickly !

[lazyest]

just to understand where I'm wrong, I've build nvidia-docker from sources, after trying to run getting this error:
nvidia-docker run --rm nvidia/cuda nvidia-smi
docker: Error response from daemon: Post http://%2Fvar%2Flib%2Fnvidia-docker%2Fnvidia-docker.sock/VolumeDriver.Mount: dial unix /var/lib/nvidia-docker/nvidia-docker.sock: connect: no such file or directory.

[lazyest]

well, I've recreate socket dir, seems to got functional one but now I'm having
nvidia-docker run --rm nvidia/cuda nvidia-smi
nvidia-docker-plugin | 2016/04/19 01:32:51 Received mount request for volume 'nvidia_driver_352.79'
nvidia-docker-plugin | 2016/04/19 01:32:51 Received mount request for volume 'nvidia_driver_352.79'
nvidia-docker-plugin | 2016/04/19 01:32:51 Received unmount request for volume 'nvidia_driver_352.79'
docker: Error response from daemon: no such volume: nvidia_driver_352.79.

this time I'm really confused :-/

[lazyest]

better result I've got is docker: Error response from daemon: no such volume: nvidia_driver_352.79.

[3XX0]

How did you install it? through the deb or using make install?
Looks like your volume didn't get created properly. Do you have a "create request" in the logs?
What's in /var/lib/nvidia-docker?

[lazyest]

I've tried both ways indeed
First I've done make deb and install it
After tried to reinstall from bundled deb to (maybe) restore config
After tried to make install
ls -l /var/lib/docker/
total 28
drwx------ 5 root root 4096 Mar 24 15:38 aufs
drwx------ 31 root root 4096 Apr 19 03:21 containers
drwx------ 3 root root 4096 Mar 24 15:38 image
drwxr-x--- 3 root root 4096 Mar 24 15:38 network
drwx------ 3 root root 4096 Apr 19 03:07 tmp
drwx------ 2 root root 4096 Mar 24 15:38 trust
drwx------ 25 nvidia-docker docker 4096 Apr 19 03:18 volumes

Maybe there is a doc about proper way of recreating volume?
it was OK before stupid docker update

ps: I didn't find exact docker logs pity :(

[3XX0]

Your installation looks really wrong, you shouldn't have anything related to nvidia-docker in /var/lib/docker/ unless you did it manually.

Try purging nvidia-docker and reinstalling the deb. By default things will be installed in /var/lib/nvidia-docker not /var/lib/docker/

sudo apt-get purge nvidia-docker
make deb
dpkg -i [...]nvidia-docker.deb

Then try to run the nvidia-smi example and check the log of the plugin /var/log/upstart/nvidia-docker.log.

If you still have issues, I suggest you open a new issue so that we can track it properly.

[lazyest]

ok