voltron/NodeMgmt
1
1
Fork 0
Code Issues Pull Requests Releases 1 Activity
Files
df4a87b1db5987858d00a4815da05b84ca367bdc
NodeMgmt/inc/status.inc
David Schroeder 67cc0da803 update
2023-11-29 15:11:22 -06:00

749 lines
32 KiB
Bash
Executable File
Raw Blame History

#!/usr/bin/env bash
STATUS(){
start=`date +%s`
PRI_CW=40;
ST_ACTION=${1}
if [ "${ST_ACTION}" = "sync" ]; then
ST_ACTION=repl
fi
declare -i cw; declare -i spc1; declare -i c
ntypesel=""
for NTYPE in "${NM_NODE_TYPES[@]}"; do
if [ "${ST_ACTION^^}" == "${NTYPE}" ]; then
ntypesel=${NTYPE}
ST_ACTION=${2}
break
fi
done
if [ "${ntypesel}" != "" ]; then
ntypesel=(${ntypesel})
else
ntypesel=(${NM_NODE_TYPES[@]})
fi
########################
if [ "${ST_ACTION}" == "report" ] || [ "${ST_ACTION}" == "repl" ] || [ "${ST_ACTION}" == "check" ] || [ "${ST_ACTION}" == "" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -en "${idsCL[LightCyan]}Setting up replication checks ... ${idsCL[Default]}"
fi
for NTYPE in "${ntypesel[@]}"; do
PH=${NTYPE}_HOSTS[0]
if [ ! -f ${NM_FOLDER}/${!PH}.down ]; then
PH_CMD="ssh root@${!PH}"
var=${NTYPE}_REPL_CHECK[@]
if [ ! -z ${!var+x} ]; then
for rcheck in "${!var}"; do
# echo "CHECK: ${NTYPE} - ${!PH} - $rcheck - ${NM_REPL_CHECK_LOC[${rcheck}]}"
${PH_CMD} rm -f ${NM_FOLDER}/test.repl
daterun=`date +%Y-%m-%d-%H-%M-%S`
${PH_CMD} "echo -e \"Status-Check (${NODE_HOSTNAME})\n${daterun}\" > ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl" &
done
fi
fi
done
# replstart=`date +%s`
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[LightGreen]}Complete${idsCL[Default]}"
echo
fi
fi
########################
if [ "${ST_ACTION}" = "" ] || [ "${ST_ACTION}" = "report" ] || [ "${ST_ACTION}" = "services" ] || [ "${ST_ACTION}" = "dockers" ] || [ "${ST_ACTION}" = "check" ]; then
if [ ! -z ${LOCAL_SERVICES+x} ] && [ "${ST_ACTION}" != "dockers" ]; then
if [ "${ST_ACTION}" != "check" ]; then
lip=${RUN_NODE_IP}
echo -e "\n$(DIVIDER)\n${idsCL[Yellow]}${idsST[Bold]} LOCALHOST Service Status${idsST[Reset]}${idsCL[Default]}"
echo -e "$(DIVIDER)\n"
echo -en " ${idsST[Bold]}${idsCL[LightCyan]}${NODE_HOSTNAME} (${lip})${idsST[Reset]}"
if ([ "${ST_ACTION}" == "report" ] && [ "${2}" == "email" ]) || [ "${ST_ACTION}" != "report" ]; then
uptime=`uptime -p`
echo -e "${idsCL[LightCyan]} - ${uptime} ${idsCL[LightYellow]}- localhost${idsCL[Default]}"
else
echo -e "${idsCL[Default]}"
fi
DIVIDER . green
fi
# if [ "${ST_ACTION}" != "check" ]; then
# lip=${RUN_NODE_IP}
# uptime=`uptime -p`
# echo -e " ${idsST[Bold]}${idsCL[LightCyan]}${NODE_HOSTNAME} (${lip})${idsST[Reset]}${idsCL[LightCyan]} - ${uptime} - localhost${idsCL[Default]}"
# DIVIDER false green
# fi
for srvc in "${LOCAL_SERVICES[@]}"; do
if [ "${ST_ACTION}" != "check" ]; then
if [ "${ST_ACTION}" != "report" ]; then
c=0; cw=${PRI_CW}; spc=''
spc1=${cw}-${#NM_SERVICE_DESC[${srvc}]}
until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
else
spc=' '
fi
echo -en "${idsCL[White]} ${NM_SERVICE_DESC[${srvc}]}${spc}${idsCL[Default]}: "
fi
if [ "$(systemctl is-active ${srvc})" != "active" ]; then
if [ ! -f ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[Red]}Not Running${idsCL[Default]}"
fi
SENDNOTICE "${NODE_HOSTNAME}" "${NM_SERVICE_DESC[${srvc}]} is down" 1
touch ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down
echo "$(date +%Y-%m-%d-%H-%M-%S) - LOCAL - ${NODE_HOSTNAME} - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${NM_LOGFILE}
else
errtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down)
if [ $errtime -gt ${NM_RENOTIFY_TIMEOUT} ]; then
if [ ! -f ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.errtime ]; then
mv ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.errtime
fi
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.errtime)
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[Red]}Not running for $(SHOW_TIME ${toterrtime})${idsCL[Default]}"
fi
touch ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down
fi
fi
else
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[LightGreen]}Running${idsCL[Default]}"
fi
if [ -f ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down ]; then
if [ -f ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.errtime ]; then
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.errtime)
else
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.down)
fi
rm -f ${NM_TMPFOLDER}/${NODE_HOSTNAME}~${srvc}.*
SENDNOTICE "${NODE_HOSTNAME}" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
echo "$(date +%Y-%m-%d-%H-%M-%S) - LOCAL - ${NODE_HOSTNAME} - ${NM_SERVICE_DESC[${srvc}]} is back up" >> ${NM_LOGFILE}
fi
fi
done
if [ "${ST_ACTION}" != "check" ]; then echo; fi
fi
########################
for NTYPE in "${ntypesel[@]}"; do
nid=1
dockers=${NTYPE}_DOCKERS_CHECK[@]
hosts=${NTYPE}_HOSTS[@]
var=${NTYPE}_HOSTS[@]
if ([[ ! -v ${NTYPE}_DOCKERS_CHECK ]] && [ "${ST_ACTION}" == "dockers" ]) || ([ "${NM_SERVICES_CHECK[${NTYPE}]}" == "" ] && [ "${ST_ACTION}" == "services" ]); then
GOFORCHECK=false;
else
GOFORCHECK=true;
fi
if [ ${GOFORCHECK} = true ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsST[Bold]}"; DIVIDER
if [ "${ST_ACTION}" == "dockers" ]; then
echo -e "${idsCL[LightYellow]} ${NM_NODETYPES[$NTYPE]}-Node Docker Status${idsCL[Default]}"
else
echo -e "${idsCL[LightYellow]} ${NM_NODETYPES[$NTYPE]}-Node Service Status${idsCL[Default]}"
fi
DIVIDER . lightYellow; echo -e "${idsST[Reset]}"
fi
for nip in "${!var}"; do
[ "${ST_ACTION}" != "check" ] && echo -en " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]} [${nip}]${idsST[Reset]}${idsCL[Default]} - Verifying Host...${idsST[Reset]}"
if [[ "${RUN_NODE_IP}" == *"${nip}"* ]]; then
NCMD=''; LH='- localhost'
else
NCMD="ssh -o ConnectTimeout=3 root@${nip}"; LH=''
fi
if [ "${NCMD}" != "" ]; then
checkhost=$(CHECK_HOST ${nip})
[ "${checkhost}" != "false" ] && checkhostssl=$(ssh -o BatchMode=yes -o ConnectTimeout=3 root@${nip} echo ok 2>&1)
else
checkhost=true
checkhostssl=ok
fi
# echo "HERE: $checkhost - $checkhostssl"
if [ "${checkhost}" != "false" ] && [ "${checkhostssl}" == "ok" ]; then
if [ "${ST_ACTION}" != "check" ]; then
if ([ "${ST_ACTION}" == "report" ] && [ "${2}" == "email" ]) || [ "${ST_ACTION}" != "report" ]; then
if [ "${NTYPE}" == "OFW" ]; then
uptime=$(${NCMD} uptime | awk '{print $3}' | cut -d, -f1)
if [ ${#uptime} -ge 6 ]; then
uptime="up $(echo $uptime | cut -d: -f1) days, $(echo $uptime | cut -d: -f2) hours, $(echo $uptime | cut -d: -f3) minutes"
elif [ ${#uptime} -ge 3 ]; then
uptime="up $(echo $uptime | cut -d: -f1) hours, $(echo $uptime | cut -d: -f2) minutes"
else
uptime="up ${uptime} minutes"
fi
else
uptime=$(${NCMD} uptime -p)
fi
[ "${ST_ACTION}" != "check" ] && echo -en "\033[K\r ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]} [${nip}]${idsST[Reset]}"
echo -e "${idsCL[White]} - ${uptime} ${idsCL[LightYellow]}${LH}${idsCL[Default]}"
else
echo -e "${idsCL[Default]}"
fi
DIVIDER false lightCyan
fi
if [ -f ${NM_TMPFOLDER}/${nip}.down ]; then
if [ -f ${NM_TMPFOLDER}/${nip}.sent ]; then
[ -f ${NM_TMPFOLDER}/${nip}.errtime ] && toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}.errtime) || toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}.down)
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]-UP" "${NM_HOSTNAMES[${nip}]}[${nip}] is back UP! It was down for $(SHOW_TIME ${toterrtime})"
fi
rm -f ${NM_TMPFOLDER}/${nip}.*
fi
########################
## SERVICES CHECK
########################
if [ "${ST_ACTION}" != "dockers" ] && [ "${NM_SERVICES_CHECK[${NTYPE}]}" != "" ]; then
echo -e "${idsCL[Green]} System Service(s) Status${idsCL[Default]}"
DIVIDER . green 52
srvcs=${NTYPE}_SERVICES_CHECK[@];
srvcstotest="$(join_by " " ${!srvcs})"
[ "${NTYPE}" != "OFW" ] && srvctst=(`${NCMD} "systemctl is-active ${srvcstotest}"`)
sr=0
for srvc in "${!srvcs}"; do
if [ "${ST_ACTION}" != "report" ]; then
c=0; cw=${PRI_CW}; spc=''
spc1=${cw}-${#NM_SERVICE_DESC[${srvc}]}
until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
else
spc=' '
fi
tmp=${NM_SINGLESRVR_SERVICES[${NTYPE}]}
if [[ ${tmp[@]} =~ ${srvc} ]]; then
if [[ $(${NCMD} ip addr show $(ip route | awk '/default/ { print $5 }') | grep "inet" | awk '/inet/ {print $2}' | cut -d'/' -f1) != *${NM_SINGLESRVR_IP[${NTYPE}]}* ]]; then
SSCHK=false
rm -f ${NM_TMPFOLDER}/${nip}~${srvc}.*
else
SSCHK=primary
fi
else
SSCHK=true
fi
[ "${srvc}" == "keepalived" ] && ([ "${nip}" == "10.2.1.2" ] || [ "${nip}" == "10.2.1.51" ]) && SSCHK=false
if [ ${SSCHK} != false ]; then
[ "${ST_ACTION}" != "check" ] && echo -en "${idsCL[White]} ${NM_SERVICE_DESC[${srvc}]}${spc}${idsCL[Default]}: "
if [ "${NTYPE}" != "OFW" ]; then
srvctest=${srvctst[$sr]}
if [ "${srvctest}" != "active" ] && [ "${srvc}" == "mysql" ]; then
[ "$(${NCMD} systemctl is-active mariadb)" == "active" ] && mysqlgo=true || mysqlgo=false
elif [ "${srvctest}" != "active" ] && [ "${srvc}" = "mariadb" ]; then
[ "$(${NCMD} systemctl is-active mysql)" == "active" ] && mysqlgo=true || mysqlgo=false
elif [ "${srvctest}" == "active" ] && ([ "${srvc}" == "mysql" ] || [ "${srvc}" == "mariadb" ]); then
mysqlgo=true
fi
else
if [ "$(ssh -o ConnectTimeout=3 root@${nip} ps -U root | grep "offsite-power-check.sh start" | grep -v "grep" | awk '{print $1}')" != "" ]; then
srvctest=active
else
srvctest=notactive
fi
fi
if [ "${srvctest}" != "active" ] && [ "${mysqlgo}" != "true" ]; then
if [ ! -f ${NM_TMPFOLDER}/${nip}~${srvc}.down ]; then
[ "${ST_ACTION}" != "check" ] && echo -en "${idsCL[Red]}Not Running"
touch ${NM_TMPFOLDER}/${nip}~${srvc}.down
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${NM_LOGFILE}
[ "${srvc}" == "pdnsadmin" ] && ${NCMD} systemctl restart ${srvc} &
elif [ ! -f ${NM_TMPFOLDER}/${nip}~${srvc}.disable ]; then
errtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${srvc}.down)
if [ $errtime -gt ${NM_RENOTIFY_TIMEOUT} ]; then
[ ! -f ${NM_TMPFOLDER}/${nip}~${srvc}.errtime ] && mv ${NM_TMPFOLDER}/${nip}~${srvc}.down ${NM_TMPFOLDER}/${nip}~${srvc}.errtime
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${srvc}.errtime)
touch ${NM_TMPFOLDER}/${nip}~${srvc}.down
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
# ([ "${srvc}" == "mysql" ] || [ "${srvc}" == "mariadb" ]) && ${NCMD} systemctl restart ${srvc} &
elif [ $errtime -gt 60 ] && [ $errtime -lt 180 ] && [ ! -f ${NM_TMPFOLDER}/${nip}~${srvc}.errtime ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
fi
[ "${ST_ACTION}" != "check" ] && echo -en "${idsCL[Red]}Not running for $(SHOW_TIME ${toterrtime})"
fi
else
[ "${ST_ACTION}" != "check" ] && echo -en "${idsCL[LightGreen]}Running"
if [ -f ${NM_TMPFOLDER}/${nip}~${srvc}.down ]; then
[ -f ${NM_TMPFOLDER}/${nip}~${srvc}.errtime ] && toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${srvc}.errtime) || toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${srvc}.down)
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is back up, it was down for $(SHOW_TIME ${toterrtime})" >> ${NM_LOGFILE}
rm -f ${NM_TMPFOLDER}/${nip}~${srvc}.*
fi
# if [ "${srvc}" == "pdnsadmin.socket" ] || [ "${srvc}" == "pdnsadmin" ] || [ "${srvc}" == "gitea" ]; then
if [ "${srvc}" == "gitea" ] || [ "${srvc}" == "headscale" ]; then
rm -f ${NM_FOLDER}/*~${srvc}.*
fi
fi
[ "${mysqlgo}" = "true" ] && unset mysqlgo
[ "${SSCHK}" == "primary" ] && echo -e "${idsCL[LightCyan]} - Primary Node${idsCL[Default]}" || echo -e "${idsCL[Default]}"
else
echo -e "${idsCL[White]} ${NM_SERVICE_DESC[${srvc}]}${spc}${idsCL[Default]}: ${idsCL[LightCyan]}Disabled - Secondary Node${idsCL[Default]}"
fi
sr=`expr $sr + 1`
done
fi
########################
## DOCKER CHECK
########################
if [[ -v ${NTYPE}_DOCKERS_CHECK ]] && [ "${ST_ACTION}" != "services" ]; then
if [ "${ST_ACTION}" != "check" ]; then
[ "${ST_ACTION}" == "" ] && [ "${NM_SERVICES_CHECK[${NTYPE}]}" != "" ] && echo
echo -e "${idsCL[Green]} Docker Service(s) Status${idsCL[Default]}"
DIVIDER . green 52
fi
for docker in "${!dockers}"; do
if [ "${ST_ACTION}" != "report" ]; then
c=0; cw=${PRI_CW}; spc=''
spc1=`expr ${cw} - ${#NM_DOCKER_DESC[${docker}]}`
until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
else
spc=' '
fi
tmp=${NM_SINGLESRVR_DOCKERS[${NTYPE}]}
if [[ ${tmp[@]} =~ ${docker} ]]; then
if [[ $(${NCMD} ip addr show $(ip route | awk '/default/ { print $5 }') | grep "inet" | awk '/inet/ {print $2}' | cut -d'/' -f1) != *${NM_SINGLESRVR_IP[${NTYPE}]}* ]]; then
SDCHK=false
rm -f ${NM_TMPFOLDER}/${nip}~${docker}.*
else
SDCHK=primary
fi
else
SDCHK=true
fi
if [ ${SDCHK} != false ]; then
[ "${ST_ACTION}" != "check" ] && echo -en "${idsCL[White]} ${NM_DOCKER_DESC[${docker}]}$spc${idsCL[Default]}: "
if [ ! "$(${NCMD} docker ps -q -f name=${docker})" ]; then
if [ "$(${NCMD} docker ps -a | grep ${docker})" ]; then
${NCMD} docker start ${docker} >/dev/null 2>&1
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is not running, starting now" >> ${NM_LOGFILE}
else
${NCMD} docker compose -f ${NM_DOCKER_COMPOSE_LOC[${docker}]}/docker-compose.yml up -d >/dev/null 2>&1
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is not found, creating and starting now" >> ${NM_LOGFILE}
fi
sleep 10s
if [ "$(${NCMD} docker ps -q -f name=${docker})" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -en "${idsCL[Green]}Running - Fixed"
fi
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_DOCKER_DESC[${docker}]} is fixed"
rm -f ${NM_TMPFOLDER}/${nip}~${docker}.*
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is fixed" >> ${NM_LOGFILE}
else
if [ "${ST_ACTION}" != "check" ]; then
echo -en "${idsCL[Red]}Not Running - Could Not Fix!"
if [ ! -f ${NM_TMPFOLDER}/${nip}~${docker}.down ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_DOCKER_DESC[${docker}]} is down, could not fix" 1
touch ${NM_TMPFOLDER}/${nip}~${docker}.down
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is down, could not fix" >> ${NM_LOGFILE}
else
errtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${docker}.down)
if [ $errtime -gt ${NM_RENOTIFY_TIMEOUT} ]; then
if [ ! -f ${NM_TMPFOLDER}/${nip}~${docker}.errtime ]; then
mv ${NM_TMPFOLDER}/${nip}~${docker}.down ${NM_TMPFOLDER}/${nip}~${docker}.errtime
fi
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}~${docker}.errtime)
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[Red]}Not running for $(SHOW_TIME ${toterrtime})${idsCL[Default]}"
fi
touch ${NM_TMPFOLDER}/${nip}~${docker}.down
fi
fi
fi
fi
else
if [ "${ST_ACTION}" != "check" ]; then
echo -en "${idsCL[LightGreen]}Running"
fi
rm -f ${NM_TMPFOLDER}/${nip}~${docker}.*
fi
[ "${SDCHK}" == "primary" ] && echo -e "${idsCL[LightCyan]} - Primary Node${idsCL[Default]}" || echo -e "${idsCL[Default]}"
else
echo -e "${idsCL[White]} ${NM_DOCKER_DESC[${docker}]}${spc}${idsCL[Default]}: ${idsCL[LightCyan]}Disabled - Secondary Node${idsCL[Default]}"
fi
done
fi
else
[ "${ST_ACTION}" != "check" ] && echo -en "\033[K\r ${idsST[Bold]}${idsCL[LightRed]}${NM_HOSTNAMES[${nip}]} [${nip}]${idsST[Reset]}"
rm -f ${NM_TMPFOLDER}/${nip}~*
if [ ! -f ${NM_TMPFOLDER}/${nip}.down ]; then
touch ${NM_TMPFOLDER}/${nip}.down
if [ ! -f ${NM_TMPFOLDER}/${nip}.errtime ]; then
touch ${NM_TMPFOLDER}/${nip}.errtime
fi
if [ "${ST_ACTION}" != "check" ]; then
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}.errtime)
if [ "${checkhost}" != "false" ]; then
echo -e "${idsCL[Red]} - Node is online, but SSL is down!${idsCL[Default]}"
else
echo -e "${idsCL[Red]} - Node is down!${idsCL[Default]}"
fi
fi
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] is down" 1
[ "${checkhostssl}" != "ok" ] && echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - Node is down" >> ${NM_LOGFILE}
else
toterrtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}.errtime)
errtime=`date +%s`-$(stat -c %Y ${NM_TMPFOLDER}/${nip}.down)
if [ "${ST_ACTION}" != "check" ]; then
if [ "${checkhost}" != "false" ]; then
echo -e "${idsCL[Red]} - Node SSL has been down for $(SHOW_TIME ${toterrtime}) ${idsCL[LightYellow]}${LH}${idsCL[Default]}"
else
echo -e "${idsCL[Red]} - Node has been down for $(SHOW_TIME ${toterrtime}) ${idsCL[LightYellow]}${LH}${idsCL[Default]}"
fi
fi
if [ $errtime -gt ${NM_RENOTIFY_TIMEOUT} ]; then
if [ "${checkhost}" != "false" ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] SSL has been down for $(SHOW_TIME ${toterrtime})" 1
else
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] has been down for $(SHOW_TIME ${toterrtime})" 1
fi
elif [ $errtime -gt 60 ] && [ $errtime -lt 180 ] && [ ! -f ${NM_TMPFOLDER}/${nip}.errtime ]; then
if [ "${checkhost}" != "false" ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] is online, but SSL is down" 1
else
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] is down" 1
fi
fi
touch ${NM_TMPFOLDER}/${nip}.down
touch ${NM_TMPFOLDER}/${nip}.sent
fi
fi
if [ "${ST_ACTION}" != "check" ]; then echo; fi
nid=`expr $nid + 1`
done
fi #GOFORCHECK
done
fi
########################
## REPLICATION CHECK
########################
if [ "${ST_ACTION}" == "report" ] || [ "${ST_ACTION}" == "repl" ] || [ "${ST_ACTION}" == "check" ] || [ "${ST_ACTION}" == "" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsST[Bold]}"; DIVIDER
echo -e "${idsCL[Yellow]} Replication Status Between the Primary and Secondary Nodes${idsCL[Default]}"
DIVIDER; echo -e "${idsST[Reset]}"
echo -en " ${idsCL[LightCyan]}Starting processes to collect/monitor replication status : "
fi
for NTYPE in "${ntypesel[@]}"; do
PH=${NTYPE}_HOSTS[0]
PH_CMD="ssh root@${!PH}"
var=${NTYPE}_REPL_CHECK[@]
if [ ! -z ${!var+x} ]; then
var=${NTYPE}_HOSTS[@]
for nip in "${!var}"; do
if [[ "${RUN_NODE_IP}" != *"${nip}"* ]] && [ "${nip}" != "${!PH}" ]; then
if [ ! -f ${NM_FOLDER}/${!PH}.down ] && [ ! -f ${NM_TMPFOLDER}/${nip}.down ]; then
var2=${NTYPE}_REPL_CHECK[@]
for rcheck in "${!var2}"; do
REPLCHECK "${rcheck}" "${nip}" "${PH_CMD}" "${ST_ACTION}" & >/dev/null 2>&1
done
fi
fi
done
fi
done
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Green]}Done${idsCL[Default]}\n"
for NTYPE in "${ntypesel[@]}"; do
PH=${NTYPE}_HOSTS[0]
PH_CMD="ssh root@${!PH}"
var=${NTYPE}_REPL_CHECK[@]
if [ ! -z ${!var+x} ]; then
var=${NTYPE}_HOSTS[@]
for nip in "${!var}"; do
if [[ "${RUN_NODE_IP}" != *"${nip}"* ]] && [ "${nip}" != "${!PH}" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -e " ${idsCL[LightCyan]}${NM_HOSTNAMES[${!PH}]} (${!PH}) <--> ${idsST[Bold]}${NM_HOSTNAMES[${nip}]}[${nip}]${idsST[Reset]}${idsCL[Default]}"
DIVIDER false green
fi
if [ ! -f ${NM_FOLDER}/${!PH}.down ] && [ ! -f ${NM_TMPFOLDER}/${nip}.down ]; then
var2=${NTYPE}_REPL_CHECK[@]
for rcheck in "${!var2}"; do
if [ "${ST_ACTION}" != "check" ]; then
if [ "${ST_ACTION}" != "report" ]; then
c=0; cw=${PRI_CW}; spc=''; spc1=${cw}-${#NM_REPL_DESC[${rcheck}]}; until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
else
spc=' '
fi
echo -en " ${NM_REPL_DESC[${rcheck}]}${spc}: "
fi
checked=""
until [ "${checked}" != "" ]; do
if [ -f ${NM_TMPFOLDER}/repl.${rcheck}.${nip}.good ]; then
checked=good
elif [ -f ${NM_TMPFOLDER}/repl.${rcheck}.${nip}.timeout ]; then
checked=timeout
fi
done
if [ "${checked}" == "timeout" ]; then
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Red]}Timeout${idsCL[Default]}"
if [ ! -f ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down ]; then
touch ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down
elif [ ! -f ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.errtime ]; then
mv ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.errtime
touch ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}[${nip}]'" "${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})" 1
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - (${NM_REPL_DESC[${rcheck}]}) Replicated folder timeout, it is not syncing" >> ${NM_LOGFILE}
else
snderrtime=`expr $(date +%s) - $(stat -c %Y ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down)`
# echo "HERE: $snderrtime"
if [ ${snderrtime} -gt ${NM_RENOTIFY_TIMEOUT} ]; then
toterrtime=`expr $(date +%s) - $(stat -c %Y ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.errtime)`
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}[${nip}]'" "${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})
It has been down for $(SHOW_TIME ${toterrtime})" 1
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - (${NM_REPL_DESC[${rcheck}]}) Replicated folder timeout, has now been down for: $(SHOW_TIME ${toterrtime})" >> ${NM_LOGFILE}
touch ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down
fi
fi
elif [ "${checked}" == "good" ]; then
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[LightGreen]}Good${idsCL[Default]}"
if [ -f ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.down ]; then
if [ -f ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.errtime ]; then
toterrtime=`expr $(date +%s) - $(stat -c %Y ${NM_TMPFOLDER}/${nip}~${rcheck}.repl.errtime)`
SENDNOTICE "Repl-'${NM_HOSTNAMES[${nip}]}[${nip}]'" "Replicated folder is back up!\n${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})
It was down for $(SHOW_TIME ${toterrtime})"
echo "$(date +%Y-%m-%d-%H-%M-%S) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - (${NM_REPL_DESC[${rcheck}]}) Replicated folder is back up" >> ${NM_LOGFILE}
fi
rm -f ${NM_TMPFOLDER}/${nip}~${rcheck}.repl*
fi
fi
done
else
if [ "${ST_ACTION}" != "check" ]; then
[ -f ${NM_FOLDER}/${!PH}.down ] && echo -e "${idsCL[Red]}${NM_HOSTNAMES[${!PH}]} (${!PH}) is offline${idsCL[Default]}"
[ -f ${NM_TMPFOLDER}/${nip}.down ] && echo -e "${idsCL[Red]}${NM_HOSTNAMES[${nip}]}[${nip}] is offline${idsCL[Default]}"
fi
fi
if [ "${ST_ACTION}" != "check" ]; then echo; fi
fi
done
fi
done
##########################
# REMOVE REPL CHECK FILES
##########################
if [ "${ST_ACTION}" = "report" ] || [ "${ST_ACTION}" = "repl" ] || [ "${ST_ACTION}" = "check" ] || [ "${ST_ACTION}" = "" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -en "${idsCL[LightCyan]} Cleaning up status checks... ${idsCL[Default]}"
fi
#NHCMD="ssh root@${NM_HOSTS['MYSQL'][0]}"
#LBHCMD="ssh root@${NM_HOSTS['LB'][0]}"
#WHCMD="ssh root@${NM_HOSTS['WEB'][0]}"
for NTYPE in "${ntypesel[@]}"; do
PH=${NTYPE}_HOSTS[0]
if [ ! -f ${NM_FOLDER}/${!PH}.down ]; then
PH_CMD="ssh root@${!PH}"
var=${NTYPE}_REPL_CHECK[@]
if [ ! -z ${!var+x} ]; then
for rcheck in "${!var}"; do
${PH_CMD} rm -f ${NM_FOLDER}/test.repl
daterun=`date +%Y-%m-%d-%H-%M-%S`
if [ "${PH_CMD}" = "" ]; then
rm -f ${NM_FOLDER}/test.repl &
rm -f ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl &
else
${PH_CMD} rm -f ${NM_FOLDER}/test.repl &
${PH_CMD} rm -f ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl &
fi
rm -f ${NM_TMPFOLDER}/repl.${rcheck}.*
done
fi
fi
done
rm -Rf ${NM_FOLDER}/test.repl
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[LightGreen]}Complete${idsCL[Default]}"
echo
fi
fi
fi
########################
## FREE SPACE CHECK
########################
#if [ "${ST_ACTION}" = "" ] || [ "${ST_ACTION}" = "report" ] || [ "${ST_ACTION}" = "freespace" ]; then
if [ "${ST_ACTION}" = "freespace" ]; then
if [ "${ST_ACTION}" != "repl" ] && [ "${ST_ACTION}" != "services" ]; then
for NTYPE in "${ntypesel[@]}"; do
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsST[Bold]}"; DIVIDER
echo -e "${idsCL[Yellow]} ${NM_NODETYPES[$NTYPE]}-Node(s) Free Space Scan${idsCL[Default]}"
DIVIDER; echo -e "${idsST[Reset]}"
fi
nid=1
var=${NTYPE}_HOSTS[@]
for nip in "${!var}"; do
if [[ $"{RNIP}" == *"${nip}"* ]]; then NCMD=''; LH='- localhost'
else NCMD="ssh -o ConnectTimeout=3 root@${nip}"; LH=''
fi
if [ "${ST_ACTION}" != "check" ]; then
echo -e " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]} [${nip}]${idsST[Reset]}${idsCL[LightCyan]} ${idsCL[LightYellow]}${LH}${idsCL[Default]}"
DIVIDER false green
fi
[ "${ST_ACTION}" != "check" ] && [ "${2}" != "report" ] && echo -en " ${idsCL[LightCyan]}Getting drives from server ... ${idsCL[Default]}"
declare -A partitions
DRIVEINFO=$(ssh -o ConnectTimeout=3 root@${nip} df -BM | grep -vE '^Filesystem|tmpfs|cdrom|@|ram|loop|udev|veeamimage|nvme|localhost|shm|mmcblk|overlay|-volume|Music|Software' | awk '{ print $1 " " $2 " " $4 }')
DRIVEINFO=(${DRIVEINFO})
if [ "${2}" != "report" ]; then
echo -en "\e[1A"
echo -e "\e[0K\r"
fi
NUMDRIVES=$((${#DRIVEINFO[@]} / 3))
for ((i = 0 ; i <= $((${NUMDRIVES}-1)) ; i++)); do
ii=$((${i}*3))
pname=`echo "${DRIVEINFO[${ii}]}" | awk -F'/' ' { print $NF } '` >/dev/null 2>&1
pname=${pname#*vg-}
if [ ${DRIVEINFO[$((${ii}+1))]//M/} -gt 1024 ]; then
freespace=${DRIVEINFO[$((${ii}+2))]//M/}
if [ "${ST_ACTION}" != "check" ]; then
c=0; cw=20; spc=''
spc1=${cw}-${#pname}
until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
echo -en " ${pname}$spc: "
fi
if [[ ${freespace} -gt 1024 ]]; then
fsgb=$(bc <<< "scale=2; ${freespace}/1024")
fsdsp="${fsgb} GB"
else
fsdsp="${freespace} MB"
fi
if [ "${freespace}" -le "1024" ]; then
fs_status='error'
fs_status_color='Red'
SENDNOTICE "Free Space Critical: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${fsdsp} free" 1
elif [ "${freespace}" -le "5120" ]; then
fs_status='warn'
fs_status_color='Yellow'
SENDNOTICE "Free Space Warning: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${fsdsp} free"
else
fs_status=''
fs_status_color='Green'
fi
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[${fs_status_color}]}${fsdsp} ${idsCL[Default]}"
fi
done
# for partition in "${partitions3[@]}"; do
# if [ "${partition}" != "udev" ] && [ "${partition}" != "/dev/sda1" ]; then
# pname=`echo "${partition}" | awk -F'/' ' { print $NF } '` >/dev/null 2>&1
# pname=${pname#*vg-}
# if [ "${ST_ACTION}" != "check" ]; then
# c=0; cw=20; spc=''
# spc1=${cw}-${#pname}
# until [ $c = ${spc1} ]; do spc="${spc} "; c=`expr $c + 1`; done
# echo -en " ${pname}$spc: "
# fi
#
# # [[ "${partition}" == *"root"* ]] && prt="/" || prt=${partition}
# # freespace=`${NCMD} df -hPBM ${prt} | awk '{print $4}' |tail -1|sed 's/M$//g'` >/dev/null 2>&1
#
# if [ "${freespace}" -le "1000" ]; then
# fs_status='error'
# fs_status_color='Red'
# SENDNOTICE "Free Space Critical: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${freespace} MB free" 1
#
# elif [ "${freespace}" -le "5000" ]; then
# fs_status='warn'
# fs_status_color='Yellow'
# SENDNOTICE "Free Space Warning: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${freespace} MB free"
#
# else
# fs_status=''
# fs_status_color='Green'
#
# fi
# if [ "${ST_ACTION}" != "check" ]; then
# if [[ ${freespace} -gt 1000 ]]; then
# fsgb=$(bc <<< "scale=2; ${freespace}/1000")
# fsdsp="${fsgb} GB"
# else
# fsdsp="${freespace} MB"
# fi
# echo -e "${idsCL[${fs_status_color}]}${fsdsp} ${idsCL[Default]}"
#
# fi
# fi
# done
nid=`expr $nid + 1`
echo
done
done
fi
fi
########################
########################
if [ "${ST_ACTION}" != "check" ]; then
echo
if [ -z ${ACTION} ] || [ "${ACTION}" = "gui" ]; then
DIVIDER true
ENTER2CONTINUE
fi
fi
end=`date +%s`
runtime=$((end-start))
echo -e "Runtime: ${runtime}\n"
echo ${runtime} >> /opt/idssys/nodemgmt/status-check.scantimes
}
REPLCHECK(){
rcheck=${1}
nip=${2}
PH_CMD=${3}
# if [ "${4}" != "check" ]; then
# timeout=`date --date='1 minutes' +%s`
# # timeout=`date --date='30 seconds' +%s`
# else
# timeout=`date --date='2 minutes' +%s`
# fi
timeout=`date --date='1 minutes' +%s`
# timeout=`date --date='30 seconds' +%s`
checked=false
until [ "${checked}" == "" ]; do
if [ "${PH_CMD}" == "" ]; then
ssh -q root@${nip} [[ -f ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl ]] && checked=`ssh -o ConnectTimeout=3 root@${nip} "cat ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl" | diff - ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl`
else
ssh -q root@${nip} [[ -f ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl ]] && checked="`${PH_CMD} \"ssh -o ConnectTimeout=3 root@${nip} \"cat ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl\" | diff - ${NM_REPL_CHECK_LOC[${rcheck}]}/test.repl\"`"
fi
[ "`date +%s`" -gt "${timeout}" ] && timeout=true && break
done
[ "${timeout}" == "true" ] && touch ${NM_TMPFOLDER}/repl.${rcheck}.${nip}.timeout || touch ${NM_TMPFOLDER}/repl.${rcheck}.${nip}.good
}
Reference in New Issue View Git Blame Copy Permalink