update check_cpu; fix typos

check_apache_requests

@@ -29,7 +29,7 @@ typeset -i iUsage=0
 # ----------------------------------------------------------------------
 # pre checks
 # ----------------------------------------------------------------------
-ph.reqire wget
+ph.require wget
 
 # ----------------------------------------------------------------------
 # check output

check_cpu

@@ -12,12 +12,51 @@
 # 2020-03-23  v1.1  <axel.hahn@iml.unibe.ch> added more data
 # 2020-07-08  v1.2  <axel.hahn@iml.unibe.ch> FIX: set "ph." instead "ps."
 # 2020-07-17  v1.3  <axel.hahn@iml.unibe.ch> use ph.require to check binaries
+# 2021-02-10  v1.4  <axel.hahn@iml.unibe.ch> added critical io wait
 # ======================================================================
 
 
 . `dirname $0`/inc_pluginfunctions
 tmpfile=/tmp/check_cpu_$$
 
+# ----------------------------------------------------------------------
+# functions
+# ----------------------------------------------------------------------
+
+function showHelp(){
+cat <<EOF
+______________________________________________________________________
+
+CHECK_CPU check cpu usage and cpu wait v1.4
+
+(c) Institute for Medical Education - Univerity of Bern
+Licence: GNU GPL 3
+______________________________________________________________________
+
+Cpu infos are taken from output of top command.
+
+
+SYNTAX:
+`basename $0` [-w WARN_LIMIT] [-c CRITICAL_LIMIT] [-i CRITICAL_IO_WAIT]
+
+OPTIONS:
+
+    -w VALUE       cpu usage warning level  (default: 75)
+    -c VALUE       cpu usage critical level (default: 90)
+
+    -i VALUE       io wait critical level   (default: 50)
+
+    -h or --help   show this help.
+
+PARAMETERS:
+
+    None.
+
+EXAMPLE:
+`basename $0` -w 60 -c 80 -i 40
+
+EOF
+}
 # ----------------------------------------------------------------------
 # MAIN
 # ----------------------------------------------------------------------
@@ -27,20 +66,19 @@ ph.require bc top
 
 
 # --- check param -h
-if [ "$1" = "-h" ]; then
-    echo "
-    usage: $0 [ -w value -c value -h ]
-
-        -w  Warning level
-        -c  Critical level
-        -h  this help
-    "
-    exit 0
-fi
+case "$1" in
+    "--help"|"-h")
+        showHelp
+        exit 0
+        ;;
+    *)
+esac
 
 # set default / override from command line params
 typeset -i iWarnLimit=`     ph.getValueWithParam 75 w "$@"`
 typeset -i iCriticalLimit=` ph.getValueWithParam 90 c "$@"`
+typeset -i iCriticalWait=`  ph.getValueWithParam 50 i "$@"`
+
 
 # get cpu status i.e.
 # %Cpu(s): 33.3 us,  9.5 sy,  0.0 ni, 57.1 id,  0.0 wa,  0.0 hi,  0.0 si,  0.0 st
@@ -66,20 +104,29 @@ cpuNonIdle=`echo 100-$cpuIdle | bc`
 
 rm -f $tmpfile
 
-if [ "$(echo "${cpuNonIdle} > ${iWarnLimit}" | bc)" -eq 1 ]; then
-
-    if [ "$(echo "${cpuNonIdle} > ${iCriticalLimit}" | bc)" -eq 1 ]; then
-        ph.setStatus "critical"
-    else
-        ph.setStatus "warning"
-    fi
+sInfo="INFO  : cpu is in normal ranges."
+if [ "$(echo "${cpuWait} > ${iCriticalWait}" | bc)" -eq 1 ]; then
+  ph.setStatus "critical"
+  sInfo="HINT  : cpu WAIT is high - check hardware issues"
+else
+  if [ "$(echo "${cpuNonIdle} > ${iWarnLimit}" | bc)" -eq 1 ]; then
+
+      if [ "$(echo "${cpuNonIdle} > ${iCriticalLimit}" | bc)" -eq 1 ]; then
+          ph.setStatus "critical"
+          sInfo="HINT  : cpu usage is high - check preocesses"
+      else
+          ph.setStatus "warning"
+          sInfo="HINT  : cpu usage is high - check preocesses"
+      fi
+  fi
 fi
 
 # --- status output
-ph.status "CPU-USAGE [%] ${cpuNonIdle} ... user: ${cpuUser} - system: ${cpuSystem} - idle: ${cpuIdle}"
-
+ph.status "CPU-USAGE [%] ${cpuNonIdle} ... user: ${cpuUser} - system: ${cpuSystem} - idle: ${cpuIdle} - wait: ${cpuWait}"
 
 echo "
+Limits: usage warn at ${iWarnLimit} .. critical at ${iCriticalLimit} .. io wait limit ${iCriticalWait}
+$sInfo
 
 Legend:
 
@@ -103,12 +150,13 @@ ph.perfadd "cpu-usage"    "${cpuNonIdle}" $iWarnLimit $iCriticalLimit 0 100
 # ph.perfadd "cpu-warn"     $iWarnLimit      "" "" 0 100
 # ph.perfadd "cpu-crit"     $iCriticalLimit  "" "" 0 100
 
+ph.perfadd "cpu-wait"     "${cpuWait}"     "" "$iCriticalWait" 0 100
+
 # --- performance data single values
 ph.perfadd "cpu-system"   "${cpuSystem}"   "" "" 0 100
 ph.perfadd "cpu-user"     "${cpuUser}"     "" "" 0 100
 ph.perfadd "cpu-idle"     "${cpuIdle}"     "" "" 0 100
 ph.perfadd "cpu-nice"     "${cpuNice}"     "" "" 0 100
-ph.perfadd "cpu-wait"     "${cpuWait}"     "" "" 0 100
 ph.perfadd "cpu-hwi"      "${cpuHi}"       "" "" 0 100
 ph.perfadd "cpu-swi"      "${cpuSi}"       "" "" 0 100
 ph.perfadd "cpu-st"       "${cpuSt}"       "" "" 0 100

check_disk-io

@@ -51,10 +51,17 @@ See https://www.kernel.org/doc/Documentation/block/stat.txt
 SYNTAX:
 `basename $0` -m MODE [-w WARN_LIMIT] [-c CRITICAL_LIMIT]
 
+OPTIONS:
+
+    -m MODE        set mode for type of output (required)
+
+    -w VALUE       warning level  (default: 0 for none)
+    -c VALUE       critical level (default: 0 for none)
+
+    -h or --help   show this help.
+
 PARAMETERS:
-    -m MODE        set mode for type of output
 
-OPTIONS:
     MODE
         io         read I/Os, write I/Os, discard I/0s
         ticks      read ticks, write ticks, discard ticks
@@ -77,26 +84,25 @@ EOF
 # echo "--- partitions: "
 # getPartitions
 
-ph.reqire bc lsblk
+ph.require bc lsblk
 
 typeset -i iDelta=0
 
 
-
-# set default / override from command line params
-typeset -i iWarnLimit=`     ph.getValueWithParam 0 w "$@"`
-typeset -i iCriticalLimit=` ph.getValueWithParam 0 c "$@"`
-sMode=`                     ph.getValueWithParam '' m "$@"`
-
-# --- set mode
 case "$1" in
-    "--help")
+    "--help"|"-h")
         showHelp
         exit 0
         ;;
     *)
 esac
 
+# set default / override from command line params
+typeset -i iWarnLimit=`     ph.getValueWithParam 0 w "$@"`
+typeset -i iCriticalLimit=` ph.getValueWithParam 0 c "$@"`
+
+# --- set mode
+sMode=`                     ph.getValueWithParam '' m "$@"`
 
 # --- labels and its columns in /sys/block/$myDisk/stat
 

check_haproxy_health

@@ -21,7 +21,7 @@ tmpfile=/tmp/check_haproxy_healthcheck_$$
 # ----------------------------------------------------------------------
 # pre checks
 # ----------------------------------------------------------------------
-ph.reqire wget
+ph.require wget
 
 cat $cfgfile >/dev/null
 if [ $? -ne 0 ]; then

check_haproxy_status

@@ -30,7 +30,7 @@ tmpfileping=/tmp/check_haproxy_status3_$$
 # ----------------------------------------------------------------------
 # pre checks
 # ----------------------------------------------------------------------
-ph.reqire wget
+ph.require wget
 
 if [ ! -f $cfgfile ]; then
   ph.abort "UNKNOWN: config file does not exist: $cfgfile"

check_memory

@@ -43,7 +43,7 @@ function getMemvalue(){
 # ----------------------------------------------------------------------
 
 # --- check required tools
-ph.reqire bc
+ph.require bc
 
 # --- check param -h
 if [ "$1" = "-h" ]; then

check_mysqlserver

@@ -190,7 +190,7 @@ if [ $bOptHelp -eq 1 -o $# -lt 1 ]; then
 fi
 
 # --- check required tools
-ph.reqire mysql
+ph.require mysql
 
 # --- install
 if [ $bOptInstall -eq 1 -a "$( whoami )" = "root" ]; then

check_netstat

@@ -19,7 +19,7 @@
 # ----------------------------------------------------------------------
 
 # --- check required tools
-ph.reqire netstat
+ph.require netstat
 
 # --- check param -h
 if [ "$1" = "-h" ]; then

check_proc_zombie

@@ -20,7 +20,7 @@ tmpfile=/tmp/check_proc_zombie_$$
 # ----------------------------------------------------------------------
 
 # --- check required tools
-ph.reqire top
+ph.require top
 
 # --- check param -h
 if [ "$1" = "-h" ]; then

check_smartstatus

@@ -5,9 +5,19 @@
 #
 # REQUIREMENTS
 # - a physical machine (no vm)
-# - sudo permissions for monitoring user to "smartctl"
+# - sudo permissions for monitoring user to "smartctl" binary
 #
 # ----------------------------------------------------------------------
+# short status in the 1st line:
+# - count of available harddisks
+# - found disks with an error
+# - each device and status
+#     - PASSED           - tests in SMART check were successful
+#     - OK               - health status OK; No Self-tests have been logged
+#     - [not supported]  - Disk available but no SMART data available (not supported)
+#     - [no access]      - disk device found but not accessible (maybe corrupt)
+#     - [does not exist] - device does not exist in the system
+# ----------------------------------------------------------------------
 #
 # ah=axel.hahn@iml.unibe.ch
 # ds=daniel.schueler@iml.unibe.ch
@@ -16,6 +26,7 @@
 # 2018-12-06  v0.4  ah,ds  show details for each disc
 # 2020-02-27  v0.5  ah,ds  add /sbin to PATH variable
 # 2020-03-05  v1.0  <axel.hahn@iml.unibe.ch> switch to ph.* helper functions
+# 2021-01-22  v1.1  <axel.hahn@iml.unibe.ch> add status [no access]
 # ======================================================================
 
 
@@ -41,8 +52,7 @@ function checkDrive(){
 
   sLabel="^SMART.*Health"
   sOK="(ok|passed)"
-
-  sOut="$sOut ${device}:"
+  sOut="$sOut | ${device}:"
 
   ls -ld $device >/dev/null 2>&1
 
@@ -53,26 +63,38 @@ function checkDrive(){
     echo >>$tmpfile
     echo "$device" >>$tmpfile
     echo >>$tmpfile
-    sudo smartctl -i $device  | grep Mode >>$tmpfile
-    echo >>$tmpfile
+    # sudo smartctl -i $device  | grep Mode >>$tmpfile
+    # echo >>$tmpfile
 
     # --- check health
-    sudo smartctl -Ha $device >>$tmpfile
-    grep -i "$sLabel" $tmpfile >/dev/null
-    if [ $? -eq 0 ]; then
-      status=`grep -i "$sLabel" $tmpfile | cut -f 2 -d ":"`
-      sOut="$sOut ${status}"
-      echo $status | egrep -i "$sOK" >>$tmpfile
-      if [ $? -ne 0 ]; then
-        iErrors=$iErrors+1
-        ph.setStatus "critical"
-      fi
+    # sudo smartctl -T permissive -Ha $device >>$tmpfile
+    sudo smartctl -Ha $device >>$tmpfile 2>&1
+    rcs=$?
+    echo $device - rc=$rcs >>$tmpfile
+    echo >>$tmpfile
+
+    if [ $rcs -eq 2 ]; then
+      iErrors=$iErrors+1
+      ph.setStatus "critical"
+      sOut="$sOut [no access]"
     else
-      sOut="$sOut [not supported]"
+      grep -i "$sLabel" $tmpfile >/dev/null
+      if [ $? -eq 0 ]; then
+        status=`grep -i "$sLabel" $tmpfile | cut -f 2 -d ":"`
+        sOut="$sOut ${status}"
+        echo $status | egrep -i "$sOK" >>$tmpfile
+        if [ $? -ne 0 ]; then
+          iErrors=$iErrors+1
+          ph.setStatus "critical"
+        fi
+      else
+        # status=`grep -i "" $tmpfile | cut -f 2 -d ":"`
+        sOut="$sOut [not supported]"
+      fi
     fi
     cat $tmpfile >>$tmpDetailsfile
-    echo >>$tmpfile
-    echo >>$tmpfile
+    # echo >>$tmpfile
+    # echo >>$tmpfile
     rm -f $tmpfile
   else
     sOut="$sOut [does not exist]"
@@ -83,7 +105,7 @@ function checkDrive(){
 # main
 # ----------------------------------------------------------------------
 
-ph.reqire smartctl
+ph.require smartctl
 
 rm -f $tmpDetailsfile 2>/dev/null
 

check_ssl

@@ -55,7 +55,7 @@ function showHelp(){
 
 # --- check requirements
 
-  ph.reqire openssl
+  ph.require openssl
 
   if [ $# -eq 0 ]; then
     showHelp

inc_pluginfunctions

@@ -55,7 +55,7 @@ function ph.abort(){
 
 # check required binaries in the path
 # param(s)  string  name of binary to check with "which" command
-function ph.reqire(){
+function ph.require(){
   which $* >/dev/null 
   if [ $? -ne 0 ]; then
       ph.setStatus "unknown"

inc_pluginfunctions.md

@@ -107,5 +107,5 @@ check if a binary PROG exists in search path (=are installed) - if not then exec
 
 Example:
 
-``ph.reqire bc lsblk``
+``ph.require bc lsblk``