Files
nrpe/files/nrpe/check_nvme_temperature
T
2026-04-24 21:30:01 +02:00

54 lines
1.5 KiB
Bash
Executable File

#!/bin/bash
# Nagios Exit Codes
STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3
# Default values
warning=70
critical=80
# Retrieve arguments
while getopts "d:w:c:" opt; do
case $opt in
d) disk=$OPTARG ;;
w) warning=$OPTARG ;;
c) critical=$OPTARG ;;
*) echo "Usage: $0 -d <disk> -w <warning> -c <critical>"; exit $STATE_UNKNOWN ;;
esac
done
# Check if disk is specified
if [ -z "$disk" ]; then
echo "ERROR: Device (e.g., /dev/nvme0) not specified."
exit $STATE_UNKNOWN
fi
# Check if smartctl is installed
if ! command -v smartctl &> /dev/null; then
echo "ERROR: smartctl is not installed."
exit $STATE_UNKNOWN
fi
# Extract temperature (looking for 'Temperature:' in smartctl output)
TEMP=$(smartctl -A "$disk" | grep -i 'Temperature:' | awk '{print $2}')
# Check if a numeric value was retrieved
if ! [[ "$TEMP" =~ ^[0-9]+$ ]]; then
echo "ERROR: Could not read temperature on $disk (check sudo permissions)."
exit $STATE_UNKNOWN
fi
# Nagios output logic
if [ "$TEMP" -ge "$critical" ]; then
echo "CRITICAL - NVMe Temperature: ${TEMP}°C (Threshold: ${critical}) | temp=${TEMP};${warning};${critical};0;100"
exit $STATE_CRITICAL
elif [ "$TEMP" -ge "$warning" ]; then
echo "WARNING - NVMe Temperature: ${TEMP}°C (Threshold: ${warning}) | temp=${TEMP};${warning};${critical};0;100"
exit $STATE_WARNING
else
echo "OK - NVMe Temperature: ${TEMP}°C | temp=${TEMP};${warning};${critical};0;100"
exit $STATE_OK
fi