voltron/NodeMgmt
1
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

update

Browse Source
This commit is contained in:
David Schroeder
2023-07-03 23:18:52 -05:00
parent c1fdda6e7d
commit 1ab1de4132
4 changed files with 69 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
inc/status.inc
View File

@@ -142,7 +142,7 @@ STATUS(){
DIVIDER; 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]}"
[ "${ST_ACTION}" != "check" ] && echo -en " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]}[${nip}]${idsST[Reset]}"
if [[ $(/sbin/ip -o -4 addr list eth0 | awk '{print $4}' | cut -d/ -f1) == *"${nip}"* ]]; then NCMD=''; LH='- localhost'
else NCMD="ssh root@${nip}"; LH=''
@@ -152,7 +152,7 @@ STATUS(){
fi
if [ "${checkhost}" != "false" ]; then
if [ "${ST_ACTION}" != "check" ]; then
#echo -en " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]} (${nip})${idsST[Reset]}"
#echo -en " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]}[${nip}]${idsST[Reset]}"
if ([ "${ST_ACTION}" == "report" ] && [ "${2}" == "email" ]) || [ "${ST_ACTION}" != "report" ]; then
uptime=`${NCMD} uptime -p`
echo -e "${idsCL[LightCyan]} - ${uptime} ${idsCL[LightYello]}${LH}${idsCL[Default]}"
@@ -169,7 +169,7 @@ STATUS(){
toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}.down)
fi
rm -f ${TMPFOLDER}/${nip}.*
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}-UP" "${NM_HOSTNAMES[${nip}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]-UP" "${NM_HOSTNAMES[${nip}]}[${nip}] is back UP! It was down for $(SHOW_TIME ${toterrtime})"
fi
if [ "${ST_ACTION}" != "dockers" ]; then
@@ -201,19 +201,19 @@ STATUS(){
if [ ! -f ${TMPFOLDER}/${nip}~${srvc}.down ]; then
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Red]}Not Running${idsCL[Default]}"
touch ${TMPFOLDER}/${nip}~${srvc}.down
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is down" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${LOGFILE}
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${LOGFILE}
else
errtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.down)
if [ $errtime -gt ${RENOTIFY} ]; then
[ ! -f ${TMPFOLDER}/${nip}~${srvc}.errtime ] && mv ${TMPFOLDER}/${nip}~${srvc}.down ${TMPFOLDER}/${nip}~${srvc}.errtime
toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.errtime)
touch ${TMPFOLDER}/${nip}~${srvc}.down
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is down" 1
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 ${TMPFOLDER}/${nip}~${srvc}.errtime ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is down" 1
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
fi
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Red]}Not running for $(SHOW_TIME ${toterrtime})${idsCL[Default]}"
@@ -222,8 +222,8 @@ STATUS(){
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Green]}Running${idsCL[Default]}"
if [ -f ${TMPFOLDER}/${nip}~${srvc}.down ]; then
[ -f ${TMPFOLDER}/${nip}~${srvc}.errtime ] && toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.errtime) || toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.down)
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_SERVICE_DESC[${srvc}]} is back up, it was down for $(SHOW_TIME ${toterrtime})" >> ${LOGFILE}
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is back up, it was down for $(SHOW_TIME ${toterrtime})" >> ${LOGFILE}
rm -f ${TMPFOLDER}/${nip}~${srvc}.*
@@ -272,9 +272,9 @@ STATUS(){
# if [ "${ST_ACTION}" != "check" ]; then
# echo -e "${idsCL[Red]}Not Running${idsCL[Default]}"
# fi
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is down" 1
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is down" 1
# touch ${TMPFOLDER}/${nip}~${srvc}.down
# echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${LOGFILE}
# echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is down" >> ${LOGFILE}
#
# else
# errtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.down)
@@ -302,8 +302,8 @@ STATUS(){
# else
# toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${srvc}.down)
# fi
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
# echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_SERVICE_DESC[${srvc}]} is back up, it was down for $(SHOW_TIME ${toterrtime})" >> ${LOGFILE}
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_SERVICE_DESC[${srvc}]} is back UP! It was down for $(SHOW_TIME ${toterrtime})"
# echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_SERVICE_DESC[${srvc}]} is back up, it was down for $(SHOW_TIME ${toterrtime})" >> ${LOGFILE}
#
# rm -f ${TMPFOLDER}/${nip}~${srvc}.down
# rm -f ${TMPFOLDER}/${nip}~${srvc}.errtime
@@ -347,28 +347,28 @@ STATUS(){
if [ ! "$(${NCMD} docker ps -q -f name=${docker})" ]; then
if [ "$(${NCMD} docker ps -a | grep ${docker})" ]; then
${NCMD} docker start ${docker} >/dev/null 2>&1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_DOCKER_DESC[${docker}]} (docker) is not running, starting now" >> ${LOGFILE}
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is not running, starting now" >> ${LOGFILE}
else
${NCMD} /usr/local/bin/docker-compose -f ${NM_DOCKER_COMPOSE_LOC[${docker}]}/docker-compose.yml up -d >/dev/null 2>&1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_DOCKER_DESC[${docker}]} (docker) is not found, creating and starting now" >> ${LOGFILE}
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is not found, creating and starting now" >> ${LOGFILE}
fi
sleep 10s
if [ "$(${NCMD} docker ps -q -f name=${docker})" ]; then
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[Green]}Running - Fixed${idsCL[Default]}"
fi
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_DOCKER_DESC[${docker}]} is fixed"
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_DOCKER_DESC[${docker}]} is fixed"
rm -f ${TMPFOLDER}/${nip}~${docker}.*
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_DOCKER_DESC[${docker}]} (docker) is fixed" >> ${LOGFILE}
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is fixed" >> ${LOGFILE}
else
if [ "${ST_ACTION}" != "check" ]; then
echo -e "${idsCL[Red]}Not Running - Could Not Fix!${idsCL[Default]}"
if [ ! -f ${TMPFOLDER}/${nip}~${docker}.down ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_DOCKER_DESC[${docker}]} is down, could not fix" 1
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_DOCKER_DESC[${docker}]} is down, could not fix" 1
touch ${TMPFOLDER}/${nip}~${docker}.down
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - ${NM_DOCKER_DESC[${docker}]} (docker) is down, could not fix" >> ${LOGFILE}
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - ${NM_DOCKER_DESC[${docker}]} (docker) is down, could not fix" >> ${LOGFILE}
else
errtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}~${docker}.down)
if [ $errtime -gt ${RENOTIFY} ]; then
@@ -405,8 +405,8 @@ STATUS(){
toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}.errtime)
echo -e "${idsCL[Red]} - Node is down!${idsCL[Default]}"
fi
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_HOSTNAMES[${nip}]} is down" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - Node is down" >> ${LOGFILE}
# SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] is down" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - Node is down" >> ${LOGFILE}
else
toterrtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}.errtime)
errtime=`date +%s`-$(stat -c %Y ${TMPFOLDER}/${nip}.down)
@@ -415,12 +415,12 @@ STATUS(){
fi
if [ $errtime -gt ${RENOTIFY} ]; then
touch ${TMPFOLDER}/${nip}.down
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_HOSTNAMES[${nip}]} has been down for $(SHOW_TIME ${toterrtime})" 1
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] has been down for $(SHOW_TIME ${toterrtime})" 1
elif [ $errtime -gt 60 ] && [ $errtime -lt 180 ] && [ ! -f ${TMPFOLDER}/${nip}.errtime ]; then
SENDNOTICE "${NM_HOSTNAMES[${nip}]}-${nip}" "${NM_HOSTNAMES[${nip}]} is down" 1
SENDNOTICE "${NM_HOSTNAMES[${nip}]}[${nip}]" "${NM_HOSTNAMES[${nip}]}[${nip}] is down" 1
fi
# if [ $errtime -gt ${RENOTIFY} ] && [ "${EMAIL_NOTICE}" != "" ]; then
# echo "${NM_HOSTNAMES[${nip}]} has been down for $(SHOW_TIME ${toterrtime})" | mail -s "${NM_HOSTNAMES[${nip}]}-${nip}" ${EMAIL_NOTICE}
# echo "${NM_HOSTNAMES[${nip}]}[${nip}] has been down for $(SHOW_TIME ${toterrtime})" | mail -s "${NM_HOSTNAMES[${nip}]}[${nip}]" ${EMAIL_NOTICE}
# fi
fi
@@ -474,7 +474,7 @@ STATUS(){
for nip in "${!var}"; do
if [[ $(/sbin/ip -o -4 addr list eth0 | awk '{print $4}' | cut -d/ -f1) != *"${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]}"
echo -e " ${idsCL[LightCyan]}${NM_HOSTNAMES[${!PH}]} (${!PH}) <--> ${idsST[Bold]}${NM_HOSTNAMES[${nip}]}[${nip}]${idsST[Reset]}${idsCL[Default]}"
DIVIDER false green
fi
if [ ! -f ${FOLDER}/${!PH}.down ] && [ ! -f ${TMPFOLDER}/${nip}.down ]; then
@@ -504,22 +504,22 @@ STATUS(){
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Red]}Timeout${idsCL[Default]}"
if [ ! -f ${TMPFOLDER}/${nip}~${rcheck}.down ]; then
touch ${TMPFOLDER}/${nip}~${rcheck}.down
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}'" "${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - (${NM_REPL_DESC[${rcheck}]}) Replicated folder timeout, it is not syncing" >> ${LOGFILE}
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}[${nip}]'" "${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})" 1
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - (${NM_REPL_DESC[${rcheck}]}) Replicated folder timeout, it is not syncing" >> ${LOGFILE}
fi
else
[ "${ST_ACTION}" != "check" ] && echo -e "${idsCL[Green]}Good${idsCL[Default]}"
if [ -f ${TMPFOLDER}/${nip}~${rcheck}.down ]; then
rm -f ${TMPFOLDER}/${nip}~${rcheck}.down
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}'" "Replicated folder is back up!\n${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})"
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]} - (${NM_REPL_DESC[${rcheck}]}) Replicated folder is back up" >> ${LOGFILE}
SENDNOTICE "Repl-Timeout-'${NM_HOSTNAMES[${nip}]}[${nip}]'" "Replicated folder is back up!\n${NM_REPL_DESC[${rcheck}]} (${NM_REPL_CHECK_LOC[${rcheck}]})"
echo "$(date) - ${nip} - ${NM_HOSTNAMES[${nip}]}[${nip}] - (${NM_REPL_DESC[${rcheck}]}) Replicated folder is back up" >> ${LOGFILE}
fi
fi
done
else
if [ "${ST_ACTION}" != "check" ]; then
[ -f ${FOLDER}/${!PH}.down ] && echo -e "${idsCL[Red]}${NM_HOSTNAMES[${!PH}]} (${!PH}) is offline${idsCL[Default]}"
[ -f ${TMPFOLDER}/${nip}.down ] && echo -e "${idsCL[Red]}${NM_HOSTNAMES[${nip}]} (${nip}) is offline${idsCL[Default]}"
[ -f ${TMPFOLDER}/${nip}.down ] && echo -e "${idsCL[Red]}${NM_HOSTNAMES[${nip}]}[${nip}] is offline${idsCL[Default]}"
fi
fi
if [ "${ST_ACTION}" != "check" ]; then echo; fi
@@ -584,7 +584,7 @@ STATUS(){
else NCMD="ssh root@${nip}"; LH=''
fi
if [ "${ST_ACTION}" != "check" ]; then
echo -e " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]} (${nip})${idsST[Reset]}${idsCL[LightCyan]} ${idsCL[LightYello]}${LH}${idsCL[Default]}"
echo -e " ${idsST[Bold]}${idsCL[LightCyan]}${NM_HOSTNAMES[${nip}]}[${nip}]${idsST[Reset]}${idsCL[LightCyan]} ${idsCL[LightYello]}${LH}${idsCL[Default]}"
DIVIDER false green
fi
@@ -617,11 +617,11 @@ STATUS(){
if [ "${freespace}" -le "1024" ]; then
fs_status='error'
fs_status_color='Red'
SENDNOTICE "Free Space Critical: '${NM_HOSTNAMES[${nip}]}'" "${partition} : ${fsdsp} free" 1
SENDNOTICE "Free Space Critical: '${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}]}'" "${partition} : ${fsdsp} free"
SENDNOTICE "Free Space Warning: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${fsdsp} free"
else
fs_status=''
fs_status_color='Green'
@@ -647,12 +647,12 @@ STATUS(){
# if [ "${freespace}" -le "1000" ]; then
# fs_status='error'
# fs_status_color='Red'
# SENDNOTICE "Free Space Critical: '${NM_HOSTNAMES[${nip}]}'" "${partition} : ${freespace} MB free" 1
# SENDNOTICE "Free Space Critical: '${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}]}'" "${partition} : ${freespace} MB free"
# SENDNOTICE "Free Space Warning: '${NM_HOSTNAMES[${nip}]}[${nip}]'" "${partition} : ${freespace} MB free"
#
# else
# fs_status=''