check_dcache_cells

#!/bin/bash
# Onno Zweers, 2011-10-06
#
# Description:
# Check if dCache cells are offline.
#

# Nagios return codes
STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3
STATE_DEPENDENT=4

CHECK_NAME="DCACHE"

PROGNAME=`basename $0`

# Check some commands that we need
/usr/bin/which lynx  2>&1 1>/dev/null || exit $STATE_UNKNOWN
/usr/bin/which perl  2>&1 1>/dev/null || exit $STATE_UNKNOWN
/usr/bin/which awk   2>&1 1>/dev/null || exit $STATE_UNKNOWN


# Process optional arguments: -c or --cell to specify cells to check.
while [ $# -gt 0 ] ; do
  case "$1" in
    -c | --cell)
      CELLS="$CELLS $2"
      shift 2
      ;;
    *)
      echo "Usage: $PROGNAME [-c cellname [-c cellname...]]" ; exit $STATE_UNKNOWN
      ;;
  esac
done


HOSTNAME=`hostname -s`
STDERR=`mktemp /tmp/$PROGNAME.tmp.XXXXXXXXXX`

# A fix for a weird bug in Lynx.
export HOME=/tmp

CELLINFO=`lynx -width=200 -dump http://dcmain.grid.sara.nl:2288/cellInfo 2> $STDERR`
ERROR=`cat $STDERR | perl -p -e 's/\s+$/ - /g' | sed -e 's/ - $//'`

# First some sanity checks
if [ -n "$ERROR" ] ; then
  echo "$CHECK_NAME UNKNOWN - $ERROR."
  exit $STATE_UNKNOWN
fi
if [ -z "$CELLINFO" ] ; then
  echo "$CHECK_NAME UNKNOWN - output of 'lynx -width=200 -dump http://dcmain.grid.sara.nl:2288/cellInfo' is empty."
  exit $STATE_UNKNOWN
fi

if [ -z "$CELLS" ] ; then
  # No cells specified: find cell list based on hostname.
  OFFLINE_CELLS=`echo "$CELLINFO" | grep ${HOSTNAME}Domain | grep    OFFLINE | awk '{print $1}' | perl -p -e 's/\s+$/; /g' | sed -e 's/; $//'`
  RUNNING_CELLS=`echo "$CELLINFO" | grep ${HOSTNAME}Domain | grep -v OFFLINE | awk '{print $1}' | perl -p -e 's/\s+$/; /g' | sed -e 's/; $//'`

  NUM_OFFLINE=`echo "$CELLINFO" | grep ${HOSTNAME}Domain | grep    OFFLINE | wc -l`
  NUM_RUNNING=`echo "$CELLINFO" | grep ${HOSTNAME}Domain | grep -v OFFLINE | wc -l`

  if [ -n "$OFFLINE_CELLS" ] ; then
    echo "$CHECK_NAME CRITICAL - $NUM_OFFLINE offline cells: $OFFLINE_CELLS"
    exit $STATE_CRITICAL
  fi
else
  # Cells specified. Check them all.
  for CELL in $CELLS ; do
    if echo "$CELLINFO" | grep --silent "$CELL" ; then
      # Cell is listed in cellinfo
      if echo "$CELLINFO" | grep "$CELL" | grep --silent OFFLINE ; then
        # Cell is offline
        NUM_OFFLINE=$(($NUM_OFFLINE + 1))
        if [ -z "$OFFLINE_CELLS" ] ; then
          OFFLINE_CELLS="$CELL"
        else
          OFFLINE_CELLS="$OFFLINE_CELLS; $CELL"
        fi
      else
        # Cell is running
        NUM_RUNNING=$(($NUM_RUNNING + 1))
        if [ -z "$RUNNING_CELLS" ] ; then
          RUNNING_CELLS="$CELL"
        else
          RUNNING_CELLS="$RUNNING_CELLS; $CELL"
        fi
      fi
    else
      # Cell is not listed!
      NUM_MISSING=$(($NUM_MISSING + 1))
      if [ -z "$MISSING_CELLS" ] ; then
        MISSING_CELLS="$CELL"
      else
        MISSING_CELLS="$MISSING_CELLS; $CELL"
      fi
    fi
  done
  # Compose report.
  if [ -n "$OFFLINE_CELLS" ] ; then
    echo "$CHECK_NAME CRITICAL - $NUM_OFFLINE offline cells: $OFFLINE_CELLS - $NUM_MISSING missing cells: $MISSING_CELLS - $NUM_RUNNING running cells: $RUNNING_CELLS"
    exit $STATE_CRITICAL
  fi
  if [ -n "$MISSING_CELLS" ] ; then
    echo "$CHECK_NAME CRITICAL - $NUM_MISSING missing cells: $MISSING_CELLS - $NUM_RUNNING running cells: $RUNNING_CELLS"
    exit $STATE_CRITICAL
  fi
fi

echo "$CHECK_NAME OK - $NUM_RUNNING running cells: $RUNNING_CELLS"
exit $STATE_OK