254 lines
10 KiB
Bash
254 lines
10 KiB
Bash
#!/bin/bash
|
|
|
|
# Monitoring plugin to check the keepalived status
|
|
usage(){
|
|
echo "Check: Is the keepalived service operate as it should.
|
|
--status | -s <state> ) Target state of the system (MASTER, BACKUP)
|
|
--interface | -i <interface> ) Interface for vrrp instance
|
|
--ha-ip | -ip <ip> ) Vrrp-ip
|
|
--interface2 | -i2 <interface> ) Interface for 2nd vrrp instance
|
|
--ha-ip2 | -ip2 <ip> ) 2nd vrrp-ip
|
|
--help | -h ) Usage
|
|
"
|
|
}
|
|
|
|
# Exit Codes
|
|
OK=0
|
|
WARNING=1
|
|
CRITICAL=2
|
|
UNKNOWN=3
|
|
|
|
# Command definitions
|
|
if [ -x "$(which cat)" ];
|
|
then
|
|
CAT="$(which cat)"
|
|
fi
|
|
|
|
if [ -x "$(which grep)" ];
|
|
then
|
|
GREP="$(which grep)"
|
|
fi
|
|
|
|
if [ -x "$(which ip)" ];
|
|
then
|
|
IP="$(which ip)"
|
|
fi
|
|
|
|
if [ -x "$(which pgrep)" ];
|
|
then
|
|
PG="$(which pgrep)"
|
|
fi
|
|
|
|
if [ -x "$(which wc)" ];
|
|
then
|
|
WC="$(which wc)"
|
|
fi
|
|
|
|
if [ "$1" = "" ]
|
|
then
|
|
echo "CRITICAL: No arguments given. Take a look at the usage:"
|
|
usage
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
# shifting through our command line arguments and setting our values
|
|
while [ "$1" != "" ]; do
|
|
case $1 in
|
|
--status | -s ) shift
|
|
TARGET_STATE="$1"
|
|
;;
|
|
--interface | -i ) shift
|
|
IFACE="$1"
|
|
;;
|
|
--ha-ip | -ip ) shift
|
|
HAIP="$1"
|
|
;;
|
|
--interface2 | -i2 ) shift
|
|
IFACE2="$1"
|
|
SEC_IP=true
|
|
;;
|
|
--ha-ip2 | -ip2 ) shift
|
|
HAIP2="$1"
|
|
;;
|
|
--help | -h ) usage
|
|
exit
|
|
;;
|
|
* ) usage
|
|
echo "CRITICAL: No valid arguments given. Take a look at the usage."
|
|
exit "${CRITICAL}"
|
|
esac
|
|
shift
|
|
done
|
|
|
|
# Config and commands
|
|
STAT_FILE='/tmp/keepalived.status'
|
|
PID_FILE='/run/keepalived.pid'
|
|
PID=$("${CAT}" "${PID_FILE}" 2>/dev/null)
|
|
SERVICE=$("${PG}" keepalived)
|
|
STATUS=(MASTER BACKUP FAULT)
|
|
CHECK_HAIP=$("${IP}" 2>/dev/null addr sh "${IFACE}" | "${GREP}" "${HAIP}" | "${WC}" -l)
|
|
CHECK_HAIP2=$("${IP}" 2>/dev/null addr sh "${IFACE2}" | "${GREP}" "${HAIP2}" | "${WC}" -l)
|
|
|
|
# Check files are valid
|
|
if [ ! -e "${STAT_FILE}" ]
|
|
then
|
|
echo "CRITICAL: Generated status file is missing. State could not be determined."
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
FILE_CONT=$("${CAT}" "${STAT_FILE}" 2>/dev/null)
|
|
|
|
if [ ! -e "${PID_FILE}" ]
|
|
then
|
|
echo "CRITICAL: PID file is missing, keepalived is not running."
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
# Check variables exists
|
|
if [ -z "${TARGET_STATE}" ]
|
|
then
|
|
echo "CRITICAL: Parameter 'status' not given. Check usage:"
|
|
usage
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
if [ -z "${IFACE}" ]
|
|
then
|
|
echo "CRITICAL: Parameter 'interface' not given. Check usage:"
|
|
usage
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
if [ -z "${HAIP}" ]
|
|
then
|
|
echo "CRITICAL: Parameter 'ha-ip' not given. Check usage:"
|
|
usage
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
# Check service is running
|
|
if [[ ! "${SERVICE}" =~ ${PID} ]]
|
|
then
|
|
echo "CRITICAL: keepalived is not running."
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
# Confirm valid STATUS
|
|
if [[ ! "${STATUS[*]}" =~ ${FILE_CONT} ]]
|
|
then
|
|
echo "CRITICAL: Status file contains unknown status or is empty. Take a look at ${STAT_FILE}."
|
|
exit "${CRITICAL}"
|
|
fi
|
|
|
|
# Check ha ip and status
|
|
if [ "${TARGET_STATE}" = "${STATUS[0]}" ] # Machine is defined as MASTER
|
|
then
|
|
case "${FILE_CONT}" in
|
|
MASTER ) STAT=ok
|
|
;;
|
|
BACKUP ) STAT=fail
|
|
;;
|
|
FAULT ) echo "CRITICAL: Machine status is FAULT."
|
|
exit "${CRITICAL}"
|
|
;;
|
|
* ) echo "CRITICAL: Status file contains unknown status or is empty. Take a look at ${STAT_FILE}."
|
|
exit "${CRITICAL}"
|
|
esac
|
|
|
|
if [ "${SEC_IP}" == "true" ] # 2nd ha instance
|
|
then
|
|
if [ "${CHECK_HAIP2}" == 0 ] && [ "${STAT}" == ok ]
|
|
then
|
|
echo "CRITICAL: 2nd HA IP ${HAIP2} is not up but machine is MASTER according to ${STAT_FILE}."
|
|
STAT_SEC_IP=CRIT
|
|
elif [ "${CHECK_HAIP2}" == 0 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: 2nd HA IP ${HAIP2} is not up, machine is BACKUP. Should be MASTER."
|
|
elif [ "${CHECK_HAIP2}" == 1 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: 2nd HA IP ${HAIP2} is up, but machine is BACKUP according to ${STAT_FILE}. Should be MASTER."
|
|
else
|
|
echo "OK: 2nd HA IP ${HAIP2} is up and machine is MASTER."
|
|
fi
|
|
fi
|
|
|
|
if [ "${CHECK_HAIP}" == 0 ] && [ "${STAT}" == ok ] # Target-actual comparison machine state; IP should be up
|
|
then
|
|
echo "CRITICAL: HA IP ${HAIP} is not up but machine is MASTER according to ${STAT_FILE}."
|
|
exit "${CRITICAL}"
|
|
elif [ "${CHECK_HAIP}" == 0 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: HA IP ${HAIP} is not up, machine is BACKUP. Should be MASTER."
|
|
exit "${CRITICAL}"
|
|
elif [ "${CHECK_HAIP}" == 1 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: HA IP ${HAIP} is up, but machine is BACKUP according to ${STAT_FILE}. Should be MASTER."
|
|
exit "${CRITICAL}"
|
|
else
|
|
echo "OK: HA IP ${HAIP} is up and machine is MASTER."
|
|
if [ "${STAT_SEC_IP}" == CRIT ]
|
|
then
|
|
exit "${CRITICAL}"
|
|
else
|
|
exit "${OK}"
|
|
fi
|
|
|
|
fi
|
|
|
|
elif [ "${TARGET_STATE}" = "${STATUS[1]}" ] # Machine is defined as BACKUP
|
|
then
|
|
case "${FILE_CONT}" in
|
|
MASTER ) STAT=fail
|
|
;;
|
|
BACKUP ) STAT=ok
|
|
;;
|
|
FAULT ) echo "CRITICAL: Machine status is FAULT."
|
|
exit "${CRITICAL}"
|
|
;;
|
|
* ) echo "CRITICAL: Status file contains unknown status or is empty."
|
|
exit "${CRITICAL}"
|
|
esac
|
|
|
|
if [ "${SEC_IP}" == "true" ] # 2nd ha instance
|
|
then
|
|
if [ "${CHECK_HAIP2}" == 1 ] && [ "${STAT}" == ok ]
|
|
then
|
|
echo "CRITICAL: 2nd HA IP ${HAIP2} is up but machine is BACKUP according to ${STAT_FILE}."
|
|
STAT_SEC_IP=CRIT
|
|
elif [ "${CHECK_HAIP2}" == 1 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "WARNING: 2nd HA IP ${HAIP2} is up, machine is MASTER. Should be BACKUP."
|
|
elif [ "${CHECK_HAIP2}" == 0 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: 2nd HA IP ${HAIP2} is not up but machine is MASTER according to ${STAT_FILE}. Should be BACKUP."
|
|
else
|
|
echo "OK: 2nd HA IP ${HAIP2} is not up and machine is BACKUP."
|
|
fi
|
|
fi
|
|
|
|
if [ "${CHECK_HAIP}" == 1 ] && [ "${STAT}" == ok ] # Target-actual comparison machine state; IP should be down
|
|
then
|
|
echo "CRITICAL: HA IP ${HAIP} is up but machine is BACKUP according to ${STAT_FILE}."
|
|
exit "${CRITICAL}"
|
|
elif [ "${CHECK_HAIP}" == 1 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "WARNING: HA IP ${HAIP} is up, machine is MASTER. Should be BACKUP."
|
|
exit "${WARNING}"
|
|
elif [ "${CHECK_HAIP}" == 0 ] && [ "${STAT}" == fail ]
|
|
then
|
|
echo "CRITICAL: HA IP ${HAIP} is not up but machine is MASTER according to ${STAT_FILE}. Should be BACKUP."
|
|
exit "${CRITICAL}"
|
|
else
|
|
echo "OK: HA IP ${HAIP} is not up and machine is BACKUP."
|
|
if [ "${STAT_SEC_IP}" == CRIT ]
|
|
then
|
|
exit "${CRITICAL}"
|
|
else
|
|
exit "${OK}"
|
|
fi
|
|
fi
|
|
|
|
else
|
|
echo "Unknown: Unknown status given. Check ${STAT_FILE} and script usage."
|
|
exit "${UNKNOWN}"
|
|
fi
|