Munin contributed plugins

#!/bin/sh

# -*- sh -*-

: << =cut

=head1 NAME

nvidia_gpu_ - Wildcard plugin to monitor NVIDIA GPUs. Uses nvidia-smi utility,

usually bundled with NVIDIA GPU driver, to obtain information.

=head1 CONFIGURATION

This is a wildcard plugin. The wildcard prefix link name should be the 

value to monitor.

This plugin uses the following configuration variables:

 [nvidia_gpu_*]

  env.smiexec - Location of nvidia-smi executable.

  env.warning - Warning temperature

  env.critical - Critical temperature

=head2 DEFAULT CONFIGURATION

The default configuration is to set "env.smiexec" to /usr/bin/nvidia-smi.

=head2 EXAMPLE WILDCARD USAGE

C<ln -s /usr/share/munin/plugins/nvidia_gpu_ /etc/munin/plugins/nvidia_gpu_temp>

...will monitor the temperature of available GPUs.

=head1 AUTHOR

Nuno Fachada

faken@fakenmc.com

=head1 LICENSE

 GNU General Public License, version 2

 http://www.gnu.org/licenses/gpl-2.0.html 

=head1 MAGIC MARKERS

 #%# family=auto

 #%# capabilities=autoconf suggest

=cut

# Determine name of parameter to monitor

name=`basename $0 | sed 's/^nvidia_gpu_//g'`

# Get location of nvidia-smi executable or use default

nvSmiExec=${smiexec:-'/usr/bin/nvidia-smi'}

# Check if autoconf was requested

if [ "$1" = "autoconf" ]; then

	# Autoconf only returns yes if nvidia-smi exists and is executable 

	if [ -x $nvSmiExec ]; then

		echo yes

		exit 0

	else

		echo "no (nvidia-smi executable not found)"

		exit 0

	fi

fi

# Check if suggest was requested

if [ "$1" = "suggest" ]; then

	echo "temp"

	echo "mem"

	echo "fan"

	exit 0

fi

# Get number of GPUs

nGpusOutput=`$nvSmiExec -L`

nGpus=`echo "$nGpusOutput" | wc -l`

if [ $nGpus -eq 0 ]; then

	# Exit if no GPUs found

	echo "No NVIDIA GPUs detected. Exiting."

	exit 1

fi

# Get full output from nvidia-smi

smiOutput=`$nvSmiExec -q`

# Check if config was requested

if [ "$1" = "config" ]; then

	# Get driver version

	driverVersion=`nvidia-smi -q | grep "Driver Version" | cut -d : -f 2 | tr -d ' '`

	# Configure graph depending on what which quantity will be plotted

	case $name in

		temp)

			echo 'graph_title GPU temperature'

			echo 'graph_args -l 0 -u 120'

			echo 'graph_vlabel Degrees (C)'

			echo 'graph_category gpu'

			echo "graph_info Temperature information for NVIDIA GPUs using driver version $driverVersion"

			nGpusCounter=0

			while [ $nGpusCounter -lt $nGpus ]

			do

				gpuName=`echo "$nGpusOutput" | sed -n $(( $nGpusCounter + 1 ))p | cut -d \( -f 1`

				echo "temp${nGpusCounter}.warning ${warning:-75}"

				echo "temp${nGpusCounter}.critical ${critical:-95}"

				echo "temp${nGpusCounter}.info Temperature information for $gpuName"

				: $(( nGpusCounter = $nGpusCounter + 1 ))

			done 

			;;

		mem)

			# First determine total memory of each GPU...

			gpusTotalMemOutput=`echo "$smiOutput" | grep -A 3 "Memory Usage" | grep "Total" | cut -d : -f 2 | tr -d ' '`

			gpusTotalMem=''

			nGpusCounter=0

			while [ $nGpusCounter -lt $nGpus ]

			do

				gpuName=`echo "$nGpusOutput" | sed -n $(( $nGpusCounter + 1 ))p | cut -d \( -f 1`

				echo "mem${nGpusCounter}.info Memory information for $gpuName"

				gpuMem=`echo "$gpusTotalMemOutput"| sed -n $(( $nGpusCounter + 1 ))p`

				gpusTotalMem="${gpusTotalMem}${gpuMem} for GPU ${nGpusCounter}"

				: $(( nGpusCounter = $nGpusCounter + 1 ))

				if [ $nGpusCounter -lt $nGpus ]; then

					gpusTotalMem="${gpusTotalMem}, "

				fi

			done

			# ...then output config data.

			echo 'graph_title GPU memory usage'

			echo 'graph_args -l 0 -u 100'

			echo 'graph_vlabel Percentage'

			echo 'graph_category gpu'

			echo "graph_info Memory usage for NVIDIA GPUs using driver version $driverVersion (total memory is $gpusTotalMem)"

			;;

		fan)

			echo 'graph_title GPU fan speed'

			echo 'graph_args -l 0 -u 100'

			echo 'graph_vlabel Percentage'

			echo 'graph_category gpu'

			echo "graph_info Fan speed of NVIDIA GPUs using driver version $driverVersion"

			nGpusCounter=0

			while [ $nGpusCounter -lt $nGpus ]

			do

				gpuName=`echo "$nGpusOutput" | sed -n $(( $nGpusCounter + 1 ))p | cut -d \( -f 1`

				echo "fan${nGpusCounter}.info Fan information for $gpuName"

				: $(( nGpusCounter = $nGpusCounter + 1 ))

				done 

			;;

		*)

			echo "Can't run without a proper symlink. Exiting."

			echo "Try running munin-node-configure --suggest."

			exit 1

			;;

	esac

	# Common stuff for all quantities

	nGpusCounter=0

	while [ $nGpusCounter -lt $nGpus ]

	do

		gpuName=`echo "$nGpusOutput" | sed -n $(( $nGpusCounter + 1 ))p | cut -d \( -f 1`

		echo "${name}${nGpusCounter}.label $gpuName"

		: $(( nGpusCounter = $nGpusCounter + 1 ))

		#print_warning $name

		#print_critical $name

	done

	exit 0

fi

# Get requested value

case $name in

	temp)

		valueGpus=`echo "$smiOutput" | grep -A 1 "Temperature" | grep "Gpu" | cut -d : -f 2 | cut -d ' ' -f 2`

		;;

	mem)

		totalMemGpus=`echo "$smiOutput" | grep -A 3 "Memory Usage" | grep "Total" | cut -d : -f 2 | cut -d ' ' -f 2`

		usedMemGpus=`echo "$smiOutput" | grep -A 3 "Memory Usage" | grep "Used" | cut -d : -f 2 | cut -d ' ' -f 2`

		valueGpus=''

		nGpusCounter=0

		while [ $nGpusCounter -lt $nGpus ]

		do

			totalMemGpu=`echo "$totalMemGpus" | sed -n $(( $nGpusCounter + 1 ))p`

			usedMemGpu=`echo "$usedMemGpus" | sed -n $(( $nGpusCounter + 1 ))p`

			percentMemUsed=$(( $usedMemGpu * 100 / $totalMemGpu ))

			valueGpus="${valueGpus}${percentMemUsed}\n"

			: $(( nGpusCounter = $nGpusCounter + 1 ))

		done

		;;

	fan)

		valueGpus=`echo "$smiOutput" | grep "Fan Speed" | cut -d ':' -f 2 | cut -d ' ' -f 2`

		;;

	*)

		echo "Can't run without a proper symlink. Exiting."

		echo "Try running munin-node-configure --suggest."

		exit 1

		;;

	esac

# Print requested value

nGpusCounter=0

while [ $nGpusCounter -lt $nGpus ]

do

	value=`echo "$valueGpus" | sed -n $(( $nGpusCounter + 1 ))p`

	echo "${name}${nGpusCounter}.value $value"

	: $(( nGpusCounter = $nGpusCounter + 1 ))

done

# TODO Follow multigraph suggestion from Flameeyes to look into multigraph plugins http://munin-monitoring.org/wiki/MultigraphSampleOutput, in order to reduce the amount of round trips to get the data.

# TODO Nvidia only: Add unsupported output options from nvidia-smi for those who have that option (how to test?). Test if they are supported and put them in suggest (or not) in case they are supported (or not)