Skip to content

ConsortiumGARR/gpu_mdev_scripts

Repository files navigation

nvidia mdevs scripts

What is it?

Scripts to save and restore VMs vgpu mdevs after reboots.

Requirement

  • properly configured nvidia drivers, passthrough and vgpu support
  • mdevctl (sudo apt-get install mdevctl)

Install

You may need to adjust SYSPATH and CREATE_PATH env variables in nvidia_mdevs_undefined_restore.sh!!!

  • Undefined mdevs file in /etc/saved_nvidia_mdevs/nvidia_mdevs is updated every 5minutes, timer starts 5minutes after system boot, you can change the frequency in save_nvidia_mdevs.timer if needed.
  • Removal of unused mdevs runs every hour, you can change it in remove_unused_nvidia_mdevs.timer if needed.

Scripts

  1. Create /etc/saved_nvidia_mdevs/ folder: sudo mkdir /etc/saved_nvidia_mdevs
  2. Copy nvidia_mdevs_saver.sh, remove_unused_nvidia_mdevs.sh and nvidia_mdevs_undefined_restore.sh to /etc/saved_nvidia_mdevs/nvidia_mdevs: sudo cp nvidia_mdevs_saver.sh remove_unused_nvidia_mkdevs.sh nvidia_mdevs_undefined_restore.sh /etc/saved_nvidia_mdevs/
  3. Set correct owner to .sh scripts: chown root:root /etc/saved_nvidia_mdevs/*.sh
  4. Set correct permissions to scripts: chmod 744 /etc/saved_nvidia_mdevs/*.sh
  5. Copy systemd units to /etc/systemd/system/: sudo cp systemd_units/*.service /etc/systemd/system/
  6. Copy systemd timers to /etc/systemd/system/sudo cp systemd_timers/*.timer /etc/systemd/system/
  7. Enable systemd units and timers:
    • systemctl enable --now save_nvidia_mdevs.timer
    • systemctl enable --now remove_unused_nvidia_mdevs.timer
    • systemctl enable restore_nvidia_mdevs.service
    • systemctl enable save_nvidia_mdevs_reboot_shutdown.service

find_nvidia_mdevs.sh

  1. Copy find_nvidia_mdevs.sh to /usr/local/sbin: cp find_nvidia_mdevs.sh /user/local/sbin/
  2. Set correct owner to .sh scripts: chown root:root /usr.local.sbin/find_nvidia_mdevs.sh
  3. Set correct permissions to scripts: chmod 744 /usr/local/sbin/find_nvidia_mdevs.sh

Example output:

root@pa1-r4-gpu01:/home/ubuntu/test_emanuele# find_nvidia_mdevs.sh 
++++++++++++++++++++++++++++++++++++++++++ Virsh VMs with nvidia mdevs +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Virsh VM Name: instance-00005a8c, OpenStack ID: 48d937b3-a626-4e7a-a6bf-3d043e74de29, mdev in use d27cb407-855c-46bd-ae6a-ac15ecfbb5d8
Virsh VM Name: instance-00005aaf, OpenStack ID: 4102db67-c94c-44a1-a94d-a6aa24ba3790, mdev in use ac3635ae-da96-4276-8998-94bfb51fcdbc
Virsh VM Name: instance-00005a8b, OpenStack ID: 888a4c61-f519-4315-abef-1050f66f4c47, mdev in use 3a961ebf-e9ed-41be-b1c5-b8a4994d74f9
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-------------------------------------------- mdevs with no associated VMs ------------------------------------------------------------
7e98b69a-57c2-4d18-b9fa-2d26db3e04c5
0cb94404-89a6-4e4c-aef3-f0d63dc3bbcb
--------------------------------------------------------------------------------------------------------------------------------------

About

Scripts to save and restore VMs vgpu mdevs after reboots

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages