Files
powerwall/offsite-power-check.sh

225 lines
10 KiB
Bash
Executable File

#!/usr/bin/env sh
PW_FOLDER='/opt/idssys/powerwall'
PW_LOGFOLDER=${PW_FOLDER}/logs
PW_LOGARCHIVE=${PW_FOLDER}/logs/archive
PW_TMPFOLDER=${PW_FOLDER}/.tmp
[ ! -d ${PW_LOGFOLDER} ] && mkdir ${PW_LOGFOLDER}
[ ! -d ${PW_LOGARCHIVE} ] && mkdir ${PW_LOGARCHIVE}
[ ! -d ${PW_TMPFOLDER} ] && mkdir ${PW_TMPFOLDER}
PW_LOGFILE=${PW_LOGFOLDER}/logfile
PW_LOGPOWER=${PW_LOGFOLDER}/log-power
[ ! -f ${PW_LOGFILE} ] && touch ${PW_LOGFILE}
[ ! -f ${PW_LOGPOWER} ] && touch ${PW_LOGPOWER}
PWO_START(){
startup=true
while true; do
. /opt/idssys/powerwall/settings.conf
PW_OFFSITEHOSTS=$(echo ${PW_OFFSITE_HOSTS} | tr ',' ' ')
# TEST=true
# TEST=full
if [ "${TEST}" != "" ]; then
PW_OFFSITE_SHUTDOWNSERVERS_AT_LOAD=95 #percent
PW_OFFSITE_SHUTDOWNSERVERS_AFTER=1 #minutes
PW_OFFSITE_WAIT_TO_POWER_SERVERS=1 #minutes
TESTMODE="### TEST MODE ACTIVE ###
"
else
TESTMODE=""
fi
upsinfo=$(upsc CP1500PFCLCD)
if [ "$(echo "${upsinfo}" | grep 'input.voltage:' | sed 's/.*: //')" != "" ]; then
rm -f ${PW_TMPFOLDER}/power.ups.losscomm
OFFSITE_VOLTIN=$(echo "${upsinfo}" | grep 'input.voltage:' | sed 's/.*: //')
OFFSITE_LOAD=$(echo "${upsinfo}" | grep 'ups.load:' | sed 's/.*: //')
OFFSITE_LOAD=`echo "scale=2; ${OFFSITE_LOAD}/100" | bc`
OFFSITE_WATTS=`echo "scale=2; ${OFFSITE_LOAD}*$(echo "${upsinfo}" | grep 'ups.realpower.nominal:' | sed 's/.*: //')" | bc`
OFFSITE_BATT_RUNTIME=$(echo "${upsinfo}" | grep 'battery.runtime:' | sed 's/.*: //')
OFFSITE_BATT_RUNTIME=`echo "scale=2; ${OFFSITE_BATT_RUNTIME}/60" | bc`
OFFSITE_BATT_CAP=$(echo "${upsinfo}" | grep 'battery.charge:' | sed 's/.*: //')
SENDNOTICE(){
MESSAGE="${TESTMODE}${2}
BATTERY CHARGE: ${OFFSITE_BATT_CAP}'%
BATT RUNTIME: ${OFFSITE_BATT_RUNTIME}'mins
AC VOLTAGE: ${OFFSITE_VOLTIN}'volts
WATT USAGE: ${OFFSITE_WATTS}'watts
$(date)"
if [ "${3}" = "1" ]; then PRIORITY="${3}"; MSGSOUND=siren; elif [ "${3}" != "" ]; then PRIORITY=${3}; else PRIORITY=0; fi
curl -m 3 -sd "token=${PW_PUSHOVER_APP_TOKEN}&user=${PW_PUSHOVER_USER_TOKEN}&message=$(echo "${MESSAGE}" | sed -e "s/%/%25/g")&title=${1}&priority=${PRIORITY}&sound=cosmic" https://api.pushover.net/1/messages.json >/dev/null 2>&1
}
if [ "${startup}" == "true" ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_WATTS}'watts - ${OFFSITE_BATT_CAP}'% Battery - ${OFFSITE_BATT_RUNTIME}'mins - Service startup" >> ${PW_LOGPOWER}
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Service startup" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE POWER MONITOR" "Service starting"
startup=false
fi
if [ ${OFFSITE_VOLTIN%.*} -gt 105 ]; then
if [ -f ${PW_TMPFOLDER}/power.offsite.charging ] && [ ${OFFSITE_BATT_CAP%.*} -eq 100 ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Charging Complete" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE POWER" "Batteries are fully recharged"
rm -f ${PW_TMPFOLDER}/power.offsite.charging
# elif [ ${OFFSITE_BATT_CAP%.*} -lt 100 ] && [ ! -f ${PW_TMPFOLDER}/power.offsite.charging ]; then
elif [ ${OFFSITE_BATT_CAP%.*} -lt 100 ] && ([ ! -f ${PW_TMPFOLDER}/power.offsite.charging ] || [ $(expr `date +%s` - $(stat -f %m ${PW_TMPFOLDER}/power.offsite.charging)) -gt 60 ]); then
# touch ${PW_TMPFOLDER}/power.offsite.charging
echo "${OFFSITE_BATT_CAP%.*}%" >| ${PW_TMPFOLDER}/power.offsite.charging
fi
if [ -f ${PW_TMPFOLDER}/power.offsite.ac.low ]; then
if [ -f ${PW_TMPFOLDER}/shutdown.OFFSITE ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Offsite Power Restored" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE POWER" "Power restored! Will wait ${PW_OFFSITE_WAIT_TO_POWER_SERVERS}'mins before turning host(s) back on"
fi
rm -f ${PW_TMPFOLDER}/power.offsite.ac.*
fi
if [ -f ${PW_TMPFOLDER}/shutdown.OFFSITE ]; then
if [ ! -f ${PW_TMPFOLDER}/power.offsite.restored ]; then
touch ${PW_TMPFOLDER}/power.offsite.restored
elif [ $(expr `date +%s` - $(stat -f %m ${PW_TMPFOLDER}/power.offsite.restored)) -gt $(echo "scale=2; ${PW_OFFSITE_WAIT_TO_POWER_SERVERS}*60" | bc) ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Powering Hosts ON After Power Outage" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE HOST POWER" "Powering offsite host(s) back ONLINE after power outage"
for host in ${PW_OFFSITEHOSTS_IDRAC}; do
if [ "${TEST}" == "" ] || [ "${TEST}" == "full" ]; then
ipmitool -I lanplus -H ${host} -U ${PW_ESXI_USER} -P "${PW_ESXI_PASS}" chassis power on &
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "sed -i '/<domain-c105105>/{n;s/<enabled>.*<\/enabled>/<enabled>True<\/enabled>/}' /etc/vmware-vpx/vpxd.cfg >/dev/null 2>&1" >/dev/null 2>&1)
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "service-control --stop vmware-vpxd >/dev/null 2>&1" >/dev/null 2>&1)
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "service-control --start vmware-vpxd >/dev/null 2>&1" >/dev/null 2>&1)
else
echo "TESTMODE: Powering up '${host}'"
fi
done
rm -f ${PW_TMPFOLDER}/power.offsite.restored
rm -f ${PW_TMPFOLDER}/shutdown.OFFSITE
fi
fi
logtime=$(expr `date +%s` - $(stat -f %m ${PW_LOGPOWER}))
if [ ${logtime} -ge $(echo "scale=0; ${PW_LOG_INTERVAL_NORMAL}*60" | bc) ] || ([ ${OFFSITE_BATT_CAP%.*} -lt 100 ] && [ ${logtime} -gt ${PW_LOG_INTERVAL_CHARGING} ]); then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_WATTS}'watts - ${OFFSITE_BATT_CAP}'% Battery - ${OFFSITE_BATT_RUNTIME}'mins - Normal" >> ${PW_LOGPOWER}
fi
else
[ ! -f ${PW_TMPFOLDER}/power.offsite.ac.low ] && touch ${PW_TMPFOLDER}/power.offsite.ac.low
if [ ! -f ${PW_TMPFOLDER}/power.offsite.ac.low.notice ] || [ $(expr `date +%s` - $(stat -f %m ${PW_TMPFOLDER}/power.offsite.ac.low.notice)) -gt $(echo "scale=2; ${PW_RENOTIFY_LOWPOWER}*60" | bc) ]; then
# echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Powering LOW/OFF!" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE POWER - POWER OFF/LOW!!" "Power off or low voltage detected" 1
touch ${PW_TMPFOLDER}/power.offsite.ac.low.notice
fi
if [ ! -f ${PW_TMPFOLDER}/shutdown.OFFSITE ] && [ ${OFFSITE_WATTS%.*} -gt 100 ] && ([ ${OFFSITE_BATT_CAP%.*} -le ${PW_OFFSITE_SHUTDOWNSERVERS_AT_LOAD} ] || [ $(expr `date +%s` - $(stat -f %m ${PW_TMPFOLDER}/power.offsite.ac.low)) -ge $(echo "scale=2; ${PW_OFFSITE_SHUTDOWNSERVERS_AFTER}*60" | bc) ]); then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_BATT_CAP}'% Battery - Powering Down Hosts!" >> ${PW_LOGFILE}
SENDNOTICE "OFFSITE HOST POWER" "Powering DOWN offsite host(s) due to power outage" 1
touch ${PW_TMPFOLDER}/shutdown.OFFSITE
for host in ${PW_OFFSITEHOSTS}; do
if [ "${TEST}" = "" ] || [ "${TEST}" = "full" ]; then
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "sed -i '/<domain-c105105>/{n;s/<enabled>.*<\/enabled>/<enabled>False<\/enabled>/}' /etc/vmware-vpx/vpxd.cfg >/dev/null 2>&1" >/dev/null 2>&1)
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "service-control --stop vmware-vpxd >/dev/null 2>&1" >/dev/null 2>&1)
# run=$(ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${PW_VCENTERHOST} "service-control --start vmware-vpxd >/dev/null 2>&1" >/dev/null 2>&1)
ssh -q -o ConnectTimeout=3 -o ConnectionAttempts=1 root@${host} sh < ${PW_FOLDER}/esxi-scripts/esxi-shutdown2.sh &
fi
done
fi
if [ $(expr `date +%s` - $(stat -f %m ${PW_LOGPOWER})) -gt ${PW_LOG_INTERVAL_ABNORMAL} ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - ${OFFSITE_VOLTIN}'volts - ${OFFSITE_WATTS}'watts - ${OFFSITE_BATT_CAP}'% Battery - ${OFFSITE_BATT_RUNTIME}'mins - OFF/LOW POWER!!" >> ${PW_LOGPOWER}
fi
fi
else
if [ ! -f ${PW_TMPFOLDER}/power.ups.losscomm ]; then
touch ${PW_TMPFOLDER}/power.ups.losscomm
elif [ $(expr `date +%s` - $(stat -f %m ${PW_TMPFOLDER}/power.ups.losscomm)) -ge 120 ]; then
echo "($(date +'%Y-%m-%d %H:%M:%S')) - 0.0'volts - 0'% Battery - Restarting NUT service" >> ${PW_LOGFILE}
SENDNOTICE "NUT DRIVER FAILURE" "Restarting NUT service"
/usr/local/etc/rc.d/nut.sh restart
sleep 10s
touch ${PW_TMPFOLDER}/power.ups.losscomm
fi
fi
sleep 5s
done
}
PWO_STOP(){
ps -U root | grep "offsite-power-check.sh start" | grep -v "grep" | awk '{print $1}' | xargs kill
}
PWO_RESTART() {
PWO_STOP
sleep 2s
/opt/idssys/powerwall/offsite-power-check.sh start &
}
LOGROTATE(){
lm=`expr $(date +'%m') - 1`
if [ ${lm} = 0 ]; then
lm=12
ly=`expr $(date +'%Y') - 1`
else
ly=$(date +'%Y')
fi
[ ${lm} -lt 10 ] && lm="0${lm}"
ARCHIVEFOLDER=${PW_LOGARCHIVE}/${ly}-${lm}
mkdir -p ${ARCHIVEFOLDER}
PWO_STOP
mv ${PW_LOGFOLDER}/log* ${ARCHIVEFOLDER}/ 2> /dev/null
PWO_START
i=0
for folder in $(ls -d ${PW_LOGARCHIVE}/* 2> /dev/null | sort -r); do
i=`expr ${i} + 1`
[ ${i} -gt 12 ] && rm -Rf ${folder}
done
}
case $1 in
start) PWO_START &;;
stop) PWO_STOP;;
restart) PWO_RESTART;;
logrotate) LOGROTATE;;
update)
if curl -m 5 -s --head --request GET https://git.schroedercity.com | grep "HTTP/2 200" > /dev/null; then
# cd /opt/idssys/defaults
# if [ "`git log --pretty=%H ...refs/heads/master^ | head -n 1`" != "`git ls-remote origin -h refs/heads/master |cut -f1`" ]; then
# git fetch origin master >/dev/null 2>&1
# git reset --hard origin/master >/dev/null 2>&1
# git reflog expire --expire=now --all >/dev/null 2>&1
# git repack -ad >/dev/null 2>&1
# git prune >/dev/null 2>&1
# git pull >/dev/null 2>&1
# git submodule update --remote >/dev/null 2>&1
# fi
cd ${PW_FOLDER}
if [ "`git log --pretty=%H ...refs/heads/master^ | head -n 1`" != "`git ls-remote origin -h refs/heads/master |cut -f1`" ]; then
git fetch origin master >/dev/null 2>&1
git reset --hard origin/master >/dev/null 2>&1
git reflog expire --expire=now --all >/dev/null 2>&1
git repack -ad >/dev/null 2>&1
git prune >/dev/null 2>&1
git pull >/dev/null 2>&1
echo "Updated, restarting services"
/opt/idssys/powerwall/offsite-power-check.sh restart &
else
echo "No update available"
fi
else
echo "Cannot connect to GIT server"
fi
;;
esac
exit 0