convey_health

#!/bin/sh
# -*- sh -*-

: << =cut

=head1 NAME

uptime - Plugin to monitor the convey's health

=head1 NOTES

Checks for the four convey outputs (anything else is treated as bad output)
  System OK (Green LED is on)
  Coprocessor Status : Successfully initialized coprocessor
  Host Sensors: OK
  CoProcessor Sensors: OK

=head1 AUTHOR

Written by Jon Daley, based on the uptime plugin

=head1 LICENSE

Public domain

=head1 MAGIC MARKERS

 #%# family=auto
 #%# capabilities=autoconf

=cut

. $MUNIN_LIBDIR/plugins/plugin.sh

CACHE_FILE=/var/run/munin/convey_health_cached

if [ "$1" = "autoconf" ]; then
	echo yes 
	exit 0
fi

if [ "$1" = "config" ]; then

	echo 'graph_title Convey Health'
	echo 'graph_args --base 1000 -l 0 '
	echo 'graph_vlabel health'
	echo 'graph_category conveyhost'
	echo 'graph_order green_led host_sensors coprocessor coprocessor_sensors coprocessor_command'

	echo 'green_led.label Green LED'
	echo 'green_led.critical 0:0'
	echo 'green_led.draw AREA'
	echo 'green_led.info 0 means OK'
	print_critical green_led
	print_warning green_led

	echo 'coprocessor.label Coprocessor Status'
	echo 'coprocessor.critical 0:0'
	echo 'coprocessor.draw STACK'
	echo 'coprocessor.info 0 means OK'
	print_critical coprocessor
	print_warning coprocessor

	echo 'host_sensors.label Host Sensors'
	echo 'host_sensors.critical 0:0'
	echo 'host_sensors.draw STACK'
	echo 'host_sensors.info 0 means OK'
	print_critical host_sensors
	print_warning host_sensors

	echo 'coprocessor_sensors.label Coprocessor Sensors'
	echo 'coprocessor_sensors.critical 0:0'
	echo 'coprocessor_sensors.draw STACK'
	echo 'coprocessor_sensors.info 0 means OK'
	print_critical coprocessor_sensors
	print_warning coprocessor_sensors

	echo 'coprocessor_command.label Coprocessor Command'
	echo 'coprocessor_command.critical 0:0'
	echo 'coprocessor_command.draw STACK'
	echo 'coprocessor_command.info 0 means OK'
	print_critical coprocessor_command
	print_warning coprocessor_command

	echo 'installed_component.label Installed Component Check'
	echo 'installed_component.critical 0:0'
	echo 'installed_component.draw STACK'
	echo 'installed_component.info 0 means OK'
	print_critical installed_component
	print_warning installed_component

	exit 0
fi
 
# the health_check script takes too long for munin, so we
# need to cache the data in a cronjob, and retrieve the data
# later
if [ "$1" = "cronjob" ]; then
  /opt/convey/support/health_check > $CACHE_FILE
  exit 0
else
  OUT=`cat $CACHE_FILE`
fi

# analyze output
echo $OUT | grep -q 'System OK (Green LED is on)'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  GREEN_LED=0
else
  GREEN_LED=1
fi

echo $OUT | grep -q 'Coprocessor Status : OK'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  COPROCESSOR=0
else
  COPROCESSOR=1
fi

echo $OUT | grep -q 'Host Sensors : OK'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  HOST_SENSORS=0
else
  HOST_SENSORS=1
fi

echo $OUT | grep -q 'Coprocessor Sensors : OK'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  COPROCESSOR_SENSORS=0
else
  COPROCESSOR_SENSORS=1
fi

echo $OUT | grep -q 'Coprocessor Command Response : OK'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  COPROCESSOR_COMMAND=0
else
  COPROCESSOR_COMMAND=1
fi

echo $OUT | grep -q 'Installed component check : OK'
FOUND=$?
if [ ${FOUND:-} == 0 ] ; then
  INSTALLED_COMPONENT=0
else
  INSTALLED_COMPONENT=1
fi

# print final output
echo "green_led.value $GREEN_LED"
echo "coprocessor.value $COPROCESSOR"
echo "host_sensors.value $HOST_SENSORS"
echo "coprocessor_sensors.value $COPROCESSOR_SENSORS"
echo "coprocessor_command.value $COPROCESSOR_COMMAND"
echo "installed_component.value $INSTALLED_COMPONENT"