From dc4cbd13a838d724578b57653ceccf1465f6a0cb Mon Sep 17 00:00:00 2001
From: Ludovic Cartier <ludovic.cartier@alterway.fr>
Date: Mon, 16 Mar 2026 15:47:32 +0100
Subject: [PATCH] add nvme check

---
 files/nrpe/check_nvme_smart       | 68 +++++++++++++++++++++++++++++++
 files/nrpe/check_nvme_temperature | 54 ++++++++++++++++++++++++
 2 files changed, 122 insertions(+)
 create mode 100644 files/nrpe/check_nvme_smart
 create mode 100644 files/nrpe/check_nvme_temperature

diff --git a/files/nrpe/check_nvme_smart b/files/nrpe/check_nvme_smart
new file mode 100644
index 0000000..f5a78e8
--- /dev/null
+++ b/files/nrpe/check_nvme_smart
@@ -0,0 +1,68 @@
+#!/bin/bash
+
+# Nagios Exit Codes
+STATE_OK=0
+STATE_WARNING=1
+STATE_CRITICAL=2
+STATE_UNKNOWN=3
+
+# Default values (in percentage)
+warning=80
+critical=90
+
+# 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_limit> -c <critical_limit>"; 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 $STATE_UNKNOWN
+fi
+
+# Retrieve SMART data (using sudo for permissions)
+SMART_DATA=$(sudo smartctl -A "$disk" 2>/dev/null)
+if [ $? -ne 0 ]; then
+    echo "UNKNOWN - Unable to run smartctl on $disk (check sudo permissions)"
+    exit $STATE_UNKNOWN
+fi
+
+# Parse the data
+USED=$(echo "$SMART_DATA" | grep "Percentage Used:" | awk '{print $3}' | tr -d '%')
+HW_WARNING=$(echo "$SMART_DATA" | grep "Critical Warning:" | awk '{print $3}')
+
+# Check if data was successfully parsed
+if [ -z "$USED" ] || [ -z "$HW_WARNING" ]; then
+    echo "UNKNOWN - Could not parse SMART data for $disk"
+    exit $STATE_UNKNOWN
+fi
+
+# Performance Data string for Nagios graphing
+PERFDATA="wear_out=${USED}%;${warning};${critical};0;100"
+
+# Evaluation Logic
+if [ "$HW_WARNING" != "0x00" ]; then
+    echo "CRITICAL - Hardware Controller Alert on $disk (Code: $HW_WARNING) | $PERFDATA"
+    exit $STATE_CRITICAL
+elif [ "$USED" -ge "$critical" ]; then
+    echo "CRITICAL - Wear level is too high: ${USED}% (Threshold: ${critical}%) | $PERFDATA"
+    exit $STATE_CRITICAL
+elif [ "$USED" -ge "$warning" ]; then
+    echo "WARNING - Wear level is increasing: ${USED}% (Threshold: ${warning}%) | $PERFDATA"
+    exit $STATE_WARNING
+else
+    echo "OK - $disk healthy (Wear: ${USED}%) | $PERFDATA"
+    exit $STATE_OK
+fi
\ No newline at end of file
diff --git a/files/nrpe/check_nvme_temperature b/files/nrpe/check_nvme_temperature
new file mode 100644
index 0000000..a5207b1
--- /dev/null
+++ b/files/nrpe/check_nvme_temperature
@@ -0,0 +1,54 @@
+#!/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
\ No newline at end of file