check_lysrdiff

#!/bin/sh
#
# This check script is maintained in a Subversion repository at
# http://lsvn.lysator.liu.se/svnroot/nagios-plugins.  Contact
# <ceder@lysator.liu.se> for commit access.

activity=26
stale=40

usage () {
  echo $0: usage: $0 '[ opt ] partno' >&2
  echo options: >&2
  echo '  --activty N  Warn if no backup completed in the last N hours.' >&2
  echo '  --stale N    Critical if oldest backup is more than N hours old.' >&2
  exit 3
}

while [ $# -gt 1 ]
do
  case "x$1" in
    x--activity)
      activity=$2
      shift	
      shift;;
    x--stale)
      stale=$2
      shift
      shift;;
    x--*)
      usage
      exit 1;;
    x*) break;;
  esac
done  

if [ $# != 1 ]
then
  usage
fi

partno=$1
BASE=/lysrdiff/$partno/perm

if [ ! -f $BASE/lysrdiff.id ]
then
    echo CRITICAL - $BASE: no such file
    exit 2
fi

if [ "`cat $BASE/lysrdiff.id`" != "$partno perm" ]
then
    echo CRITICAL - found `cat $BASE/lysrdiff.id` instead of $lysrdiff >&2
    exit 2
fi

acttime=`date '+%Y-%m-%d %H:%M:%S' -d "$activity hours ago"`
staletime=`date '+%Y-%m-%d %H:%M:%S' -d "$stale hours ago"`

statecache=/tmp/lysrdiff-nagios-state
state=$statecache-$partno

find $BASE/lysrdiff/state -type f \
	-printf "%TY-%Tm-%Td %TH:%TM:%TS %P\n" \
      | sort > $statecache.$partno.$$
mv -f $statecache.$partno.$$ $state || {
  echo UNKNOWN - mv failed
  rm $statecache.$partno.$$
  exit 3
}
rm -f $state-stale $state-active || {
  echo UNKNOWN - rm failed
  rm $state
  exit 3
}

echo $staletime MARKER \
| sort $state - \
| grep -- '-start$\|MARKER' \
| sed -e '/MARKER$/,$d' \
> $state-stale

echo $acttime MARKER \
| sort $state - \
| grep -- '-end$\|MARKER' \
| sed -e '0,/MARKER$/d' \
> $state-active

FAILED=`grep -c -- '-fail$' $state`
STALE=`wc -l < $state-stale`
ACTIVITY=`wc -l < $state-active`

if [ $FAILED -gt 0 ]
then
  echo -n "CRITICAL - $FAILED failed backups"
  if [ $STALE -gt 0 ]
  then
      echo -n " (also $STALE stale)"
  fi
  echo -n ": "
  sed -n 's/-fail$//p' $state \
    |awk 'BEGIN { sz = 0 }
	  NR > 1 && sz < 198 { printf ", "; sz += 2 }
	  sz < 200 { printf "%s", $3; sz += length($3) }
	  END { if (sz >= 200) printf ", others" } ' \
    |sed s'/, \([^,]*\)$/ and \1/'
  echo '. | activity='$ACTIVITY stale=$STALE failed=$FAILED
  rm $state $state-stale $state-active

  exit 2
fi

if [ $STALE -gt 0 ]
then
  echo -n "CRITICAL - $STALE stale backups: "
  sed 's/-start$//' $state-stale \
    |awk 'BEGIN { sz = 0 }
	  NR > 1 && sz < 198 { printf ", "; sz += 2 }
	  sz < 200 { printf "%s", $3; sz += length($3) }
	  END { if (sz >= 200) printf ", others" } ' \
    |sed s'/, \([^,]*\)$/ and \1/'
  echo '. | activity='$ACTIVITY stale=$STALE failed=0
  rm $state $state-stale $state-active
  exit 2
fi

if [ $ACTIVITY = 0 ]
then
  echo 'WARNING - No activity. | activity='0 stale=$STALE failed=0
  rm $state $state-stale $state-active
  exit 1
fi

echo 'OK - all ok. | activity='$ACTIVITY stale=0 failed=0
rm $state $state-stale $state-active
exit 0