Proxmox script to update nvidia driver on system and in containers

nvidia-driver-update.sh

#!/bin/bash

function usage() {
  echo "Usage: $0 -u DRIVERURL -c CTID0,CTID1,..." 1>&2; exit 0;
}

while getopts ":hu:c:" opt; do
  case "$opt" in
    u) driverurl="$OPTARG";;
    c) containers="$OPTARG";;
    h | *) usage;;
  esac
done

if [ -z "${driverurl}" ]; then
        usage
fi

function logmsg() {
  echo "######## $1 ########"
}

tmpfile=/tmp/nvidia.run
installcmd="$tmpfile --ui=none --no-questions"



# update system
function update_system_driver() {
  logmsg "Updating system driver"
  eval $installcmd
  logmsg "Finished system driver install"
}



# update container
function update_container_driver() {

  ct=$1
  shift

  # Initial container status
  status=$(pct list | grep $ct | tr -s ' ' | cut -f2 -d' ')
  if [ "$status" = "stopped" ]; then
    pct start $ct
  fi

  logmsg "Updating driver in $ct"
  logmsg "mounting $ct"
  pct mount $ct
  logmsg "copying $tmpfile to container"
  cp $tmpfile /var/lib/lxc/$ct/rootfs$tmpfile

  pct exec $ct -- bash -c "$@ --no-kernel-module"
  rm /var/lib/lxc/$ct/rootfs$tmpfile
  pct unmount $ct
  logmsg "Finished in $ct"

  pct stop $ct

  # Start container if it was originally running
  if [ "$status" = "running" ]; then
    pct start $ct
    logmsg "$ct was originally running, so restarting"
  fi
}


# download the file
logmsg "Downloading driver from $driverurl to $tmpfile"
curl -o $tmpfile $driverurl
chmod +x $tmpfile


### Prep the system

## kill services in running containers
for i in ${containers//,/ }
do
  # Stop possible GPU processes
  pct exec $i -- bash -c "systemctl stop xxnetwork-cmix"
done

# update system driver
update_system_driver

# step through each container and update
for i in ${containers//,/ }
do
  update_container_driver $i "$installcmd"
done



### Clean up the system
logmsg "Deleting tempfile $tmpfile"
rm $tmpfile