Adding wear leveling changes to check_smart. Adding check_systemd and check_zfs

Imanol-Mikel Barba Sabariego
1 parent 75ead3b5
Showing 6 changed files with 222 additions and 15 deletions
check_smart/auxiliar.cpp
check_smart/auxiliar.h
check_smart/check_smart.cpp
check_smart/check_smart.h
check_systemd/check_systemd.sh
check_zfs/check_zfs.sh
@@ -52,4 +52,4 @@ int exec(string cmd, string *output)
         }
     }
     return pclose(pipe)/256;
-}
+}
 \ No newline at end of file
 #include "check_smart.h"
+#include <regex>
 const char *servicename = (const char*)"SMART";
-int getSmartAttrValue(string line) {
-  return stoi(line.substr(line.find_last_of(" ")+1));
+int getSmartAttrValue(string line, unsigned int col) {
+  line = std::regex_replace(line, std::regex("\\s+"), " ");
+  line = std::regex_replace(line, std::regex("^ "), "");
+
+  // Find nth col
+  size_t pos = 0;
+  int  cnt = 0;
+
+  while(cnt != col) {
+    pos = line.find(" ", pos+1);
+    if (pos == std::string::npos) {
+      throw std::runtime_error("Column out of range");
+    }
+    cnt++;
+  }
+  if(pos != 0) {
+    ++pos;
+  }
+
+  size_t end_pos = line.find(" ", pos);
+  if(end_pos != std::string::npos) {
+     return stoi(line.substr(pos, line.find(" ", pos)));
+  }
+  return stoi(line.substr(pos));
 }
 int getSmartAttrID(string line) {
@@ -88,7 +111,7 @@ int evalStatus(const char* disk, int driveType, string *status) {
         continue;
       }
       if(getSmartAttrID(line) == id) {
-        attrMap[id].value = getSmartAttrValue(line);
+        attrMap[id].value = getSmartAttrValue(line, attrMap[id].col);
       }
     }
   }
@@ -107,11 +130,21 @@ int evalStatus(const char* disk, int driveType, string *status) {
     }
     int veredict = 0;
-    if(attr.value > attr.threshold_warn) {
-      veredict = WARN;
-    }
-    if(attr.threshold_crit != -1 && attr.value > attr.threshold_crit) {
-      veredict = CRIT;
+
+    if(attr.lower_than) {
+      if(attr.value < attr.threshold_warn) {
+        veredict = WARN;
+      }
+      if(attr.threshold_crit != -1 && attr.value < attr.threshold_crit) {
+        veredict = CRIT;
+      }
+    } else {
+      if(attr.value > attr.threshold_warn) {
+        veredict = WARN;
+      }
+      if(attr.threshold_crit != -1 && attr.value > attr.threshold_crit) {
+        veredict = CRIT;
+      }
     }
     switch(veredict) {
@@ -183,7 +216,7 @@ int main(int argc, char **argv) {
     switch(c) {
       case 'h':
         printHelp(true);
-	return OK;
+        return OK;
       case 'V':
         printVersion();
         return OK;
@@ -48,6 +48,8 @@ struct SMARTAttr
   int threshold_warn;
   int threshold_crit;
   bool optional;
+  unsigned int col;
+  bool lower_than;
 }; typedef struct SMARTAttr SMARTAttr;
@@ -59,6 +61,8 @@ SMARTAttr reallocated = {
   .threshold_warn = 0,
   .threshold_crit = -1,
   .optional = false,
+  .col = 9,
+  .lower_than = false,
 };
 SMARTAttr pending = {
@@ -66,8 +70,10 @@ SMARTAttr pending = {
   .name = "Current_Pending_Sector",
   .value = -1,
   .threshold_warn = 0,
-  .threshold_crit = -1, 
+  .threshold_crit = -1,
   .optional = false,
+  .col = 9,
+  .lower_than = false,
 };
 SMARTAttr off_uncorrect = {
@@ -77,24 +83,30 @@ SMARTAttr off_uncorrect = {
   .threshold_warn = 0,
   .threshold_crit = 0, 
   .optional = false,
+  .col = 9,
+  .lower_than = false,
 };
 SMARTAttr wear = {
   .id = WEAR_COUNT_ID,
   .name = "Wear_Leveling_Count",
   .value = -1,
-  .threshold_warn = 80,
-  .threshold_crit = 90, 
+  .threshold_warn = 20,
+  .threshold_crit = 10, 
   .optional = true,
+  .col = 3,
+  .lower_than = true,
 };
 SMARTAttr wearout = {
   .id = MEDIA_WEAROUT_ID,
   .name = "Media_Wearout_Indicator",
   .value = -1,
-  .threshold_warn = 80,
-  .threshold_crit = 90,
+  .threshold_warn = 20,
+  .threshold_crit = 10,
   .optional = true,
+  .col = 3,
+  .lower_than = true,
 };
 SMARTAttr badblocks = {
@@ -104,6 +116,8 @@ SMARTAttr badblocks = {
   .threshold_warn = 0,
   .threshold_crit = 0, 
   .optional = false,
+  .col = 9,
+  .lower_than = false,
 };
 SMARTAttr rep_uncorrect = {
@@ -113,6 +127,8 @@ SMARTAttr rep_uncorrect = {
   .threshold_warn = 0,
   .threshold_crit = -1, 
   .optional = false,
+  .col = 9,
+  .lower_than = false,
 };
 map<int,SMARTAttr> prepareAttrMap(int driveType);
+#!/bin/bash
+
+UNIT=$1
+
+systemctl status $UNIT > /dev/null
+RET=$?
+
+if [[ $RET != 0 ]]; then
+  echo "SERVICE $UNIT - CRITICAL: Not running"
+  exit 2
+fi
+
+echo "SERVICE $UNIT - OK"
+exit 0
+#!/bin/bash
+
+function smart_disks() {
+  if [[ $# != 1 ]]; then
+    echo "Wrong number of arguments"
+    return 3
+  fi
+
+  POOL=$1
+  DISKS=$(zpool status -P $POOL | sed -E 's/\t/ /g' | sed -E 's/[ ]+/ /g' | sed -E 's/^ //g' | grep "^/dev" | cut -d' ' -f 1)
+  ERR_OUTPUT=""
+  HIGHEST_RES=0
+  for disk in $DISKS; do
+    OUTPUT=$(/usr/lib64/nagios/plugins/check_smart $disk)
+    RES=$?
+    if [[ $RES != 0 ]]; then
+      ERR_OUTPUT="$ERR_OUTPUT - $OUTPUT"
+    fi
+    if [[ $RES -gt $HIGHEST_RES ]]; then
+      HIGHEST_RES=$RES
+    fi
+  done
+
+  echo -n "ZFS POOL SMART $POOL "
+  if [[ $HIGHEST_RES == 1 ]]; then
+    echo "WARNING${ERR_OUTPUT}"
+    return 1
+  elif [[ $HIGHEST_RES == 2 ]]; then
+    echo "CRITICAL${ERR_OUTPUT}"
+    return 2
+  elif [[ $HIGHEST_RES == 3 ]]; then
+    echo "UNKNOWN${ERR_OUTPUT}"
+    return 3
+  fi
+
+  echo "OK"
+  return 0
+}
+
+function pool_errors() {
+  if [[ $# != 1 ]]; then
+    echo "Wrong number of arguments"
+    return 3
+  fi
+
+  OUTPUT="$(zpool status $1)"
+  if [[ $? != 0 ]]; then
+    return 3
+  fi
+
+  STATUS=$(echo "$OUTPUT" | grep -P "^\s*state:" | sed 's/ //g' | grep -P "^\s*state" | cut -d ':' -f 2)
+  if [[ $? != 0 ]]; then
+    return 3
+  fi
+
+  ERRORS=$(echo "$OUTPUT" | grep -P "^errors:")
+  if [[ $? != 0 ]]; then
+    return 3
+  fi
+
+  if [[ $STATUS != "ONLINE" ]]; then
+    echo "ZFS POOL ERRORS $1 - CRITICAL: status: $STATUS"
+    return 2
+  fi
+
+  if [[ $ERRORS != "errors: No known data errors" ]]; then
+    echo "ZFS POOL ERRORS $1 - WARNING: $ERRORS"
+    return 1
+  fi
+
+  echo "ZFS POOL ERRORS $1 - OK"
+  return 0
+}
+
+function pool_free() {
+  if [[ $# != 7 ]]; then
+    echo "Wrong number of arguments"
+    return 3
+  fi
+  POOL=$3
+
+  FREE_RAW=$(zfs list -po available $POOL | tail -n 1)
+  if [[ $? != 0 ]]; then
+    return 3
+  fi
+
+  USED_RAW=$(zfs list -po used $POOL | tail -n 1)
+  if [[ $? != 0 ]]; then
+    return 3
+  fi
+  TOTAL_RAW=$((USED_RAW + FREE_RAW))
+  TOTAL=$(echo "scale=2; $TOTAL_RAW / (1024^3)" | bc)
+  FREE=$(echo "scale=2; $FREE_RAW / (1024^3)" | bc)
+  PERC_FREE=$(echo "scale=2; 100*$FREE_RAW/$TOTAL_RAW" | bc)
+
+  CRITICAL="0"
+  WARNING="0"
+  TEMP=$(getopt -o c:w: -- "$@")
+  eval set -- "$TEMP"
+
+  for opt; do
+    case "$opt" in
+        -c) CRITICAL=$2; shift 2 ;;
+        -w) WARNING=$2;  shift 2 ;;
+    esac
+  done
+
+  if [[ $CRITICAL == "0" || $WARNING == "0" ]]; then
+    echo "Missing -c or -w arguments"
+    return 3
+  fi
+
+  if [[ $(echo "$PERC_FREE < $CRITICAL" | bc) == 1 ]]; then
+    echo "ZFS POOL FREE $POOL - CRITICAL: ${PERC_FREE}% ${FREE}/${TOTAL} GB"
+    return 2
+  elif [[ $(echo "$PERC_FREE < $WARNING" | bc) == 1 ]]; then
+    echo "ZFS POOL FREE $POOL - WARNING: ${PERC_FREE}% ${FREE}/${TOTAL} GB"
+    return 2
+  fi
+
+  echo "ZFS POOL FREE $POOL - OK: ${PERC_FREE}% ${FREE}/${TOTAL} GB"
+  return 0
+}
+
+if [[ $# -lt 2 ]]; then
+  echo "Wrong number of arguments"
+  exit 3
+fi
+
+ACTION=$1
+POOL=$2
+
+if [[ $ACTION == "smart" ]]; then
+  smart_disks $POOL
+  exit $?
+elif [[ $ACTION == "errors" ]]; then
+  pool_errors $POOL
+  exit $?
+elif [[ $ACTION == "free" ]]; then
+  pool_free $POOL $@
+  exit $?
+fi
+echo "Unknown command: $ACTION"
+exit 3