MiniKF VMs - Fail on Startup with Error
This is a public announcement from Arrikto informing the MiniKF community members that starting from 4 August 2022, MiniKF VMs both new & existing ones fail on startup with an error.
NVIDIA released an update to the nvidia-kernel-common-510 package, which is not built properly by NVIDIA and as a result cannot get installed via unattended-upgrades. This causes the following error with MiniKF:
An unexpected exception has occurred
Reason: Command `<ExtCommand [uSVxpH1Y6XI] `apt-get install --yes --no-install-recommends google-cloud-sdk=335.0.0-0', status=FINISHED (ret: 100), PID=22483, shell=False>' failed. Error log: E: Unmet dependencies. Try 'apt --fix-broken install' with no packages (or specify a solution).\n
The logfile `/home/minikf/provision.log' may contain more information on this error, including a backtrace.
*** MiniKF provisioning tool terminated. Press Enter to restart, Ctrl-C to exit... ***
To keep the previous desired behavior, users have to manually disable unattended-upgrades and pin a specific NVIDIA version, following the procedure described below:
1. Make sure you are in the MiniKF setup screen, with the above error message.
2. Press Ctrl+B, then c . You will see a new prompt.
3. Stop unattended-upgrades:
minikf@minikf:~$ sudo systemctl stop unattended-upgrades.service
4. Pin NVIDIA CUDA to a specific version:
minikf@minikf:~$ cat << EOF | sudo tee /etc/apt/preferences.d/nvidia.pref
Package: /nvidia/ /cuda-drivers/
Pin: version 510.73.08*
minikf@minikf:~$ sudo apt-get install -f --yes --allow-downgrades
minikf@minikf:~$ sudo apt-get remove unattended-upgrades --yes
7. Reboot your MiniKF:
minikf@minikf:~$ sudo reboot
Verify that MiniKF is able to boot.
Arrikto will be pushing a new version of MiniKF that will take this fix into account. Check back or follow this KB article to get notified when Arrikto pushes the new marketplace images to all of the support public clouds.