packages/nagios/skel/etc/init.d/nagios

   1 #!/bin/bash
   2
   3 # chkconfig: 345 99 01
   4 # description: Nagios network monitoring daemon
   5
   6 ### BEGIN INIT INFO
   7 # Provides:       nagios
   8 # Required-Start:
   9 # Required-Stop:
  10 # Default-Start:  2 3 5
  11 # Default-Stop:
  12 # Description:    Nagios network monitoring daemon
  13 ### END INIT INFO
  14
  15 # Author: Lars Michelsen <lm@mathias-kettner.de>
  16
  17 # Notes for OMD init script requirements
  18 # - Must handle omd config options like daemon enabling/disabling
  19 # - When a daemon is disabled by omd config it needs
  20 #   to return an exit code of 5.
  21 # - The init script must output an exit code of 2 when
  22 #   an unknown param is used.
  23 # - In general the exit code for succeeded actions is
  24 #   0 and for failed actions it is 1.
  25 # - There are exceptions for the exit code handling:
  26 #   - When a service is already stopped and should be
  27 #     restarted/stopped, it should result in an exit code of 0.
  28 #   - When a service is already running and should be started
  29 #     this also should result in an exit code of 0.
  30 # - When a restart is requested and the program is still not running
  31 #   the script should only execute a start
  32 # - When a restart is requested and the program can not be stopped the
  33 #   script should terminate without starting the daemon
  34 # - When a reload is requested and the program is not running
  35 #   the init script should execute a start instead
  36
  37 cd ###ROOT###
  38 . .profile
  39 . lib/omd/init_profile
  40 . etc/omd/site.conf
  41 [ "$CONFIG_CORE" = "nagios" ] || exit 5
  42
  43 BIN=###ROOT###/bin/nagios
  44 CFG_FILE=###ROOT###/tmp/nagios/nagios.cfg
  45 STATUS_FILE=###ROOT###/tmp/nagios/status.dat
  46 CMD_FILE=###ROOT###/tmp/run/nagios.cmd
  47 PID_FILE=###ROOT###/tmp/lock/nagios.lock
  48 STATUS_DAT=###ROOT###/tmp/nagios/status.dat
  49 CHECKRESULTS_DIR=###ROOT###/tmp/nagios/checkresults
  50 USR=###SITE###
  51 GRP=###SITE###
  52
  53 # You can set the environment variable CORE_NOVERIFY=yes
  54 # in order to supress a verification of the nagios configuration
  55 # in case of start, restart or reload. This is in order to
  56 # avoid duplicate effort when being called by cmk -R or
  57 # cmk -O.
  58 # export CORE_NOVERIFY=yes
  59
  60 # Make sure that check plugins do not run localized.
  61 # check_icmp outputs performance data with german
  62 # comma instead of dot and makes it unparsable.
  63 unset LANG
  64 export LC_ALL=C
  65
  66 # OMD: create configuration file out of fragments
  67 case "$1" in start|restart|reload|checkconfig|check)
  68     merge-nagios-config \
  69        ###ROOT###/etc/nagios/nagios.d/*.cfg \
  70        ###ROOT###/etc/nagios/nagios.cfg \
  71        > $CFG_FILE || rm -f $CFG_FILE
  72     if [ $? -ne 0 ]; then exit 1; fi
  73 esac
  74
  75 OPTIONS="-ud"
  76
  77 # Fetches the pid of the currently running nagios process of the given
  78 # user.
  79 #
  80 # --ppid 1 in ps seem not to filter by direct ppid but by the whole
  81 # parent process tree. So filter by hand again.
  82 #
  83 # Removed the filter "-P 1" (filters for ppid=1 processes) as on some
  84 # distros, like Ubuntu 13.10 and newer, the processes will not be childs
  85 # of PID 1, instead the process is child of an "upstart user session",
  86 # which is visible via ps as "init --user". This will be the PPID until
  87 # the user session ends, then the process will be moved to PPID=1.
  88 # Strange one, but we try to simply ignore that...  "-o" should make it.
  89 #
  90 # It returns 1 when no process can be found and echos the PID while
  91 # returning 0 when a process can be found.
  92 pidof_nagios() {
  93     pgrep -u $OMD_SITE -o -fx "$BIN $OPTIONS $CFG_FILE" 2>/dev/null
  94 }
  95
  96
  97 verify_config() {
  98     if [ "$1" != "quiet" ]; then
  99         echo -n "Running configuration check... "
 100     fi
 101     RESULT=$($BIN -pv $CFG_FILE 2>&1)
 102     if [ $? -eq 0 ]; then
 103         if [ "$1" != "quiet" ]; then
 104             echo "done."
 105             echo "$RESULT" >&2
 106         fi
 107         return 0
 108     else
 109         if [ "$1" != "quiet" ]; then
 110             echo "CONFIG ERROR! Aborted. Check your Nagios configuration."
 111         fi
 112         echo "$RESULT" >&2
 113         return 1
 114     fi
 115 }
 116
 117 prep_start() {
 118     if [ -f $CMD_FILE ]; then
 119         rm -f $CMD_FILE
 120     fi
 121     touch $PID_FILE
 122     chown $USR:$GRP $PID_FILE
 123     rm -f $CHECKRESULTS_DIR/*
 124 }
 125
 126 nagios_wait_stop() {
 127     pid=$(pidof_nagios) || {
 128         echo -n 'not running...'
 129         return 0
 130     }
 131
 132     # wait until really stopped.
 133     # it might happen that nagios has a subprocess which
 134     # is left running and becomes ppid 1 after killing the
 135     # main nagios process. So fetch the process id again
 136     # multiple times to fetch new processes until all are gone.
 137     I=0
 138     while kill -0 $pid >/dev/null 2>&1; do
 139         # Send TERM to process group to kill the nagios process and also
 140         # other processes started by this nagios process, for example
 141         # check plugins which are currently running
 142         kill -TERM -$pid
 143         while kill -0 $pid  >/dev/null 2>&1;  do
 144             if [ $I = '60' ]; then
 145                 echo -ne "\nsending SIGKILL"
 146                 kill -9 $pid
 147             elif [ $I = '70' ]; then
 148                 return 1
 149             fi
 150
 151             echo -n "."
 152             I=$(($I+1))
 153             sleep 1
 154         done
 155         # Is there another proc with ppid 1?
 156         pid=$(pidof_nagios) || break
 157     done
 158
 159     rm -f "$PID_FILE"
 160 }
 161
 162 nagios_wait_start() {
 163     prep_start
 164     $BIN $OPTIONS $CFG_FILE
 165
 166     I=0
 167     while ! pidof_nagios >/dev/null 2>&1;  do
 168         if [ $I = '10' ]; then
 169             return 1
 170         else
 171             echo -n "."
 172             I=$(($I+1))
 173             sleep 1
 174         fi
 175     done
 176 }
 177
 178 if [ ! -f $BIN ]; then
 179     echo "Nagios binary $BIN not found. Terminating..."
 180     exit 1
 181 fi
 182
 183 case "$1" in start|restart|reload|checkconfig)
 184     if [ ! -f $CFG_FILE ]; then
 185         echo "Nagios configuration file $CFG_FILE not found. Terminating..."
 186         exit 1
 187     fi
 188 esac
 189
 190 __init_hook $0 $1 pre
 191 case "$1" in
 192     start)
 193          echo -n "Starting nagios..."
 194          if pidof_nagios >/dev/null 2>&1; then
 195              echo 'Already running.'
 196              exit 0
 197          fi
 198
 199          [ "$CORE_NOVERIFY" ] || verify_config quiet || exit 1
 200
 201          if nagios_wait_start; then
 202              echo 'OK'
 203              __init_hook $0 $1 post 0
 204              exit 0
 205          else
 206              echo 'ERROR'
 207              __init_hook $0 $1 post 1
 208              exit 1
 209          fi
 210     ;;
 211     stop)
 212         echo -n "Stopping nagios..."
 213         if nagios_wait_stop; then
 214             echo 'OK'
 215              __init_hook $0 $1 post 0
 216             exit 0
 217         else
 218             echo 'ERROR'
 219              __init_hook $0 $1 post 1
 220             exit 1
 221         fi
 222     ;;
 223     check|checkconfig)
 224         if ! verify_config; then
 225             exit 1
 226         fi
 227         exit 0
 228     ;;
 229     status)
 230         if pid=$(pidof_nagios 2>&1)
 231         then
 232             echo "Running ($pid)."
 233         else
 234             echo 'Not running.'
 235             exit 1
 236         fi
 237     ;;
 238     restart)
 239         [ "$CORE_NOVERIFY" ] || verify_config quiet || exit 1
 240
 241         $0 stop || (echo "Unable to stop nagios. Terminating..." && exit 1)
 242         echo -n "Starting nagios..."
 243         if nagios_wait_start; then
 244             echo 'OK'
 245             exit 0
 246         else
 247             echo 'ERROR'
 248             exit 1
 249         fi
 250     ;;
 251
 252     reload|force-reload)
 253         [ "$CORE_NOVERIFY" ] || verify_config quiet || exit 1
 254
 255         # Execute a start when nagios is not running
 256         if ! pid=$(pidof_nagios) 2>&1; then
 257             $0 start
 258             exit $?
 259         fi
 260
 261         echo -n "Reloading nagios configuration (PID: $pid)... "
 262         if kill -HUP $pid >/dev/null 2>&1; then
 263             echo 'OK'
 264             __init_hook $0 $1 post 0
 265             exit 0
 266         else
 267             $0 restart
 268             exit $?
 269         fi
 270     ;;
 271
 272     *)
 273         echo "Usage: nagios {start|stop|restart|reload|status|checkconfig}"
 274         exit 2
 275     ;;
 276 esac
 277
 278 # EOF