nagios service dependence handler 0.2

#!/bin/bash -xu

SERVICESTATE=""
LASTSERVICESTATE=""
HOSTNAME=""
SERVICEDESC=""
DEPENDENTS=""
CRITICALSERVICE=""
DEPINDX=0

OK=0
WARNING=1
CRITICAL=2
UNKNOWN=3

COMMANDFILE="/var/nagios/rw/nagios.cmd"
LIVE="/var/nagios/rw/live"

usage() {
    echo $0 ' -c $SERVICESTATE$ -l $LASTSERVICESTATE$ -d DEPENDENTS -H $HOSTNAME$ -s $SERVICEDESC$ -o SOMEOUTPUT'
}

while getopts "c:l:d:H:S:o:Ch" OPT; do
    case $OPT in
    c)
        SERVICESTATE=$OPTARG
        ;;
    l)
        LASTSERVICESTATE=$OPTARG
        ;;
    d)
        DEPENDENTS=($(echo $OPTARG| tr -s ',' ' '))
        ;;
    H)
        HOSTNAME=$OPTARG
        ;;
    S)
        SERVICEDESC=$OPTARG
        ;;
    o)
        OUTPUT=$OPTARG
        ;;
    C)
        CRITICALSERVICE="TRUE"
        ;;
    h)
        usage
        exit 0
        ;;
    ?)
        usage
        exit 1
        ;;
    esac
done

if test -z "$SERVICESTATE" -o -z "$LASTSERVICESTATE" -o -z "${DEPENDENTS[*]}" -o -z "$HOSTNAME" -o -z "$SERVICEDESC"; then
    usage
    exit 1
fi

_get_service_plugin_output() {
    echo "GET services
Filter: description = $1
Columns: plugin_output" | unixcat $LIVE
}

_get_service_state() {
    echo "GET services
Filter: description = $1
Columns: state" | unixcat $LIVE
}

_get_service_error_counters() {
    if test -z "$1"; then
    echo "_get_service_error_counters called without arguments"
    exit 1
    fi

    local PLUGINOUTPUT=$(_get_service_plugin_output $1)
    local CRIT_COUNTER=${PLUGINOUTPUT##*_FAIL_}
    local WARN_COUNTER=${PLUGINOUTPUT%%_FAIL_*}

    # WARN_COUNTER is empty or is not a number
    if [[ -z "$WARN_COUNTER" || ! "$WARN_COUNTER" =~ "^[0-9]+$" ]]; then
    WARN_COUNTER=0
    fi

    if [[ -z "$CRIT_COUNTER" || ! "$CRIT_COUNTER" =~ "^[0-9]+$" ]]; then
    CRIT_COUNTER=0
    fi

    echo "$WARN_COUNTER $CRIT_COUNTER"
}

# $1 => The service to receive the unkown state
# $2 => The plugin_output
# $3 => Set the status to critical if $3 is true
_change_state() {
    local SERVICE=$1
    local OUTPUT=$2
    local STATUS=$3
    echo "[$(date +%s)] PROCESS_SERVICE_CHECK_RESULT;$HOSTNAME;$SERVICE;$STATUS;$OUTPUT" > $COMMANDFILE
}

# $1 => The service to enable active checks
_enable_active_checks() {
    echo "[$(date +%s)] ENABLE_SVC_CHECK;$HOSTNAME;$1" > $COMMANDFILE
}

# $1 => The service to diable active checks
_disable_active_checks() {
    echo "[$(date +%s)] DISABLE_SVC_CHECK;$HOSTNAME;$1" > $COMMANDFILE
}


ok_to_other() {
    for DEP in ${DEPENDENTS[*]}; do

    # Get error counter
    local ERRORCOUNTERS=($(_get_service_error_counters "$DEP"))
    local WARN_COUNTER=${ERRORCOUNTERS[0]}
    local CRIT_COUNTER=${ERRORCOUNTERS[1]}

        # Disble active checks on dependents, if is not already disabled
    if test "$WARN_COUNTER" -eq "0" -a "$CRIT_COUNTER" -eq "0"; then
        _disable_active_checks $DEP
    fi


        # Passive change the state
    local DEPSTATE=$(_get_service_state $DEP)
    if test -n "$CRITICALSERVICE" -o "$DEPSTATE" -eq "$CRITICAL"; then
        let CRIT_COUNTER++
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $CRITICAL # Change/Keep to CRITICAL
    else
        let WARN_COUNTER++
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $WARNING # Change to WARNING
    fi
    done
}

other_to_ok() {
    for DEP in ${DEPENDENTS[*]}; do

    local ERRORCOUNTERS=($(_get_service_error_counters "$DEP"))
    local WARN_COUNTER=${ERRORCOUNTERS[0]}
    local CRIT_COUNTER=${ERRORCOUNTERS[1]}

    local DEPSTATE=$(_get_service_state $DEP)

    if test -n "$CRITICALSERVICE"; then
        let CRIT_COUNTER--
        if test "$CRIT_COUNTER" -eq "0" -a "$WARN_COUNTER" -gt "0"; then
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $WARNING
        elif test "$CRIT_COUNTER" -gt "0"; then
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $CRITICAL
        elif test "$CRIT_COUNTER" -lt "0"; then
        _change_state $DEP ERROR $UNKNOWN
        fi
    else
        let WARN_COUNTER--
        if test "$DEPSTATE" == "$CRITICAL"; then
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $CRITICAL
        elif test "$WARN_COUNTER" -ge "0"; then
        _change_state $DEP ${WARN_COUNTER}_FAIL_${CRIT_COUNTER} $WARNING
        else
        _change_state $DEP ERROR $UNKNOWN
        fi
    fi

    # If error counter is 0 enable active checks
    if test "$WARN_COUNTER" -eq "0" -a "$CRIT_COUNTER" -eq "0"; then
        _enable_active_checks $DEP
    fi
    done
}

if test "$LASTSERVICESTATE" == "OK" -a "$SERVICESTATE" != "OK"; then
    ok_to_other
elif test "$LASTSERVICESTATE" != "OK" -a "$SERVICESTATE" == "OK"; then
    other_to_ok
fi