Update dsmon.sh

This commit is contained in:
2024-01-06 23:19:07 -06:00
parent 3a37c5e96a
commit 8a88fa1378

302
dsmon.sh
View File

@@ -127,150 +127,178 @@ CHECK(){
#for hostid in "${!host_ip[@]}"; do
for hostname in "${hostnames_sort[@]}"; do
hostid=${host_ids[$hostname]}
# if [ "${1}" != "report" ]; then
echo -e "${idsST[Bold]}"; DIVIDER false lightCyan 85
echo -e "${idsCL[LightCyan]} ${host_name[$hostid]} - Drive Space Check${idsCL[Default]}"
DIVIDER false lightCyan 85
# fi
checkhost=$(CHECK_HOST ${host_ip[$hostid]})
if [ "${checkhost}" != "false" ]; then
declare -A host_limits_tmp
hostlimits=(${host_limits[${hostid}]})
for hl in ${hostlimits[@]}; do
hlname=$(echo $hl | cut -d ":" -f1)
hllim=$(echo $hl | cut -d ":" -f2)
host_limits_tmp[$hlname]=$hllim
done
DRIVEINFO=$(ssh root@${host_ip[$hostid]} 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})
# for x in "${!DRIVEINFO[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO[$x]}" ; done
NUMDRIVES=$((${#DRIVEINFO[@]} / 3))
declare -A DRIVEINFO_TOT DRIVEINFO_FREE DRIVEINFO_FREEPER DRIVEINFO_SHORTNAME
declare -a DRIVES
for ((i = 0 ; i <= $((${NUMDRIVES}-1)) ; i++)); do
ii=$((${i}*3))
dname=${DRIVEINFO[${ii}]}
dname=${dname//\/dev\/mapper\//}
dname=${dname//\/dev\//}
dname_short=${dname#*vg-}
dname_short=${dname_short%*_v}
[ "$dname_short" = "" ] && dname_short=$dname
if [ "${dname}" = "sda1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda14" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda15" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda2" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "shm" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "nvme0n1p1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "mmcblk0p1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
else
dtot=$(bc <<< "scale=2; ${DRIVEINFO[$((${ii}+1))]}/1024/10")
dfree=$(bc <<< "scale=2; ${DRIVEINFO[$((${ii}+2))]}/1024/10")
dfreeper=$(printf "%0.2f" $(jq -n ${dfree}/${dtot}*100))
DRIVEINFO_TOT[$dname]=$dtot
DRIVEINFO_FREE[$dname]=$dfree
DRIVEINFO_FREEPER[$dname]=$dfreeper
DRIVEINFO_SHORTNAME[$dname]=$dname_short
DRIVES+=("$dname")
fi
done
IFS=$'\n' DRIVES=($(sort <<<"${DRIVES[*]}")); unset IFS
# for x in "${!DRIVEINFO_TOT[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_TOT[$x]}" ; done
# echo
# for x in "${!DRIVEINFO_FREE[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_FREE[$x]}" ; done
# echo
# for x in "${!DRIVEINFO_FREEPER[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_FREEPER[$x]}" ; done
# echo
# if [ "${1}" != "report" ]; then
c=0; cw=18; spcA=''; spc1=`expr ${cw} - 7`; until [ $c = ${spc1} ]; do spcA="${spcA} "; c=`expr $c + 1`; done
c=0; cw=10; spcB=''; spc2=`expr ${cw} - 4`; until [ $c = ${spc2} ]; do spcB="${spcB} "; c=`expr $c + 1`; done
c=0; cw=10; spcC=''; spc3=`expr ${cw} - 5`; until [ $c = ${spc3} ]; do spcC="${spcC} "; c=`expr $c + 1`; done
c=0; cw=14; spcD=''; spc4=`expr ${cw} - 3`; until [ $c = ${spc4} ]; do spcD="${spcD} "; c=`expr $c + 1`; done
echo -e "${idsST[Reset]}${idsCL[White]} DRIVE${spcA} FREE GB${spcB}FREE %${spcC}TOTAL GB${spcD}WARN % / CRIT %${idsCL[Default]}"
# pf="%-20s %10s %10s %12s %50s\n"
# printf "${pf}" "${idsCL[White]}DRIVE" "FREE GB" "FREE %" "TOTAL GB" "WARN % / CRIT %${idsCL[Default]}"
# fi
for DRIVE in "${DRIVES[@]}"; do
[ "${host_limits_tmp[$DRIVE]}" = "" ] && host_limits_tmp[$DRIVE]=81
WARNING_LEVEL=$(printf "%0.2f" $(jq -n 100-${host_limits_tmp[$DRIVE]}))
CRITICAL_LEVEL=$(printf "%0.2f" $(jq -n $WARNING_LEVEL-$WARNING_LEVEL/2))
# if [ "${1}" != "report" ]; then
c=0; cw=18; spcA=''; spc1=`expr ${cw} - ${#DRIVEINFO_SHORTNAME[$DRIVE]}`; until [ $c = ${spc1} ]; do spcA="${spcA} "; c=`expr $c + 1`; done
c=0; cw=10; spcB=''; spc2=`expr ${cw} - ${#DRIVEINFO_FREE[$DRIVE]}`; until [ $c = ${spc2} ]; do spcB="${spcB} "; c=`expr $c + 1`; done
c=0; cw=10; spcC=''; spc3=`expr ${cw} - ${#DRIVEINFO_FREEPER[$DRIVE]}`; until [ $c = ${spc3} ]; do spcC="${spcC} "; c=`expr $c + 1`; done
c=0; cw=14; spcD=''; spc4=`expr ${cw} - ${#DRIVEINFO_TOT[$DRIVE]}`; until [ $c = ${spc4} ]; do spcD="${spcD} "; c=`expr $c + 1`; done
# fi
if (( $(bc <<<"${DRIVEINFO_FREEPER[$DRIVE]} <= ${CRITICAL_LEVEL}") )); then
fs_status='CRITICAL'
fs_priority=2
fs_status_color='RedBold'
elif (( $(bc <<<"${DRIVEINFO_FREEPER[$DRIVE]} <= ${WARNING_LEVEL}") )); then
fs_status='Warning'
fs_priority=1
fs_status_color='Yellow'
elif (( $(bc <<<"${DRIVEINFO_FREE[$DRIVE]} < 5") )); then
fs_status='Warning'
fs_priority=1
fs_status_color='Magenta'
else
fs_status='Normal'
fs_priority=0
fs_status_color='Green'
fi
D_WARNING_LEVEL=$(printf "%0.2f" $(jq -n 100-${WARNING_LEVEL}))
D_CRITICAL_LEVEL=$(printf "%0.2f" $(jq -n 100-${CRITICAL_LEVEL}))
# [ "${1}" != "report" ] &&
echo -e "${idsCL[Cyan]} ${DRIVEINFO_SHORTNAME[$DRIVE]}${spcA}${idsCL[$fs_status_color]}${DRIVEINFO_FREE[$DRIVE]} GB${spcB}${DRIVEINFO_FREEPER[$DRIVE]}%${spcC}${DRIVEINFO_TOT[$DRIVE]} GB${idsST[Reset]}${spcD}${idsCL[Cyan]}( ${idsCL[Yellow]}${D_WARNING_LEVEL}%${idsCL[Cyan]} / ${idsCL[RedBold]}${D_CRITICAL_LEVEL}%${idsST[Reset]}${idsCL[Cyan]} )${idsCL[Default]}"
# [ "${1}" != "report" ] && printf "${pf}" "${idsCL[Cyan]}${DRIVEINFO_SHORTNAME[$DRIVE]}" "${idsCL[$fs_status_color]}${DRIVEINFO_FREE[$DRIVE]} GB" "${DRIVEINFO_FREEPER[$DRIVE]}%" "${DRIVEINFO_TOT[$DRIVE]} GB" "${idsCL[Cyan]}(${idsCL[Default]}${D_WARNING_LEVEL}%${idsCL[Cyan]}/${idsCL[Default]}${D_CRITICAL_LEVEL}%${idsCL[Cyan]})(${idsCL[Yellow]}${D_WARNING_LEVEL}%${idsCL[Cyan]}/${idsCL[Red]}${D_CRITICAL_LEVEL}%${idsCL[Cyan]})${idsCL[Default]}"
if [ "${fs_status}" != "Normal" ] && [ "${1}" == "report" ]; then
# echo -e "${host_name[$hostid]} - (${host_ip[$hostid]})\n\n${DRIVE} : ${DRIVEINFO_FREE[$DRIVE]}GB out of ${DRIVEINFO_TOT[$DRIVE]}GB Free (${DRIVEINFO_FREEPER[$DRIVE]}%)\n\n$(date)" | mail -s "Free Space ${fs_status}: '${host_name[$hostid]}'" ${ALERT_EMAIL}
SENDNOTICE "Free Space ${fs_status}: '${host_name[$hostid]}'" "${host_name[$hostid]} - (${host_ip[$hostid]})
${DRIVE} : ${DRIVEINFO_FREE[$DRIVE]}GB out of ${DRIVEINFO_TOT[$DRIVE]}GB Free (${DRIVEINFO_FREEPER[$DRIVE]})percent" ${fs_priority}
fi
done
unset DRIVEINFO_TOT DRIVEINFO_FREE DRIVEINFO_FREEPER DRIVEINFO_SHORTNAME DRIVES DRIVEINFO host_limits_tmp
else
# [ "${1}" != "report" ] &&
echo -e "${idsCL[Yellow]} ${host_name[$hostid]} is down${idsCL[Default]}"; echo
RUN_CHECK ${hostname} > /tmp/dscheck.${hostname}.results 2>/dev/null &
done
nc_count=0; completed=false; scanstart=$(date +%s)
until [ "${completed}" == "true" ]; do
for host in /tmp/dscheck.*.done; do
NTS=$(grep -oP '(?<=dscheck.).*?(?=.done)' <<< "${nodestatus}")
if [ "${NTS}" != "*" ]; then
hoststatus=$(cat /tmp/dscheck.${NTS}.results)
[ "${hoststatus}" != "" ] && echo -e "${hoststatus}"
rm -f /tmp/dscheck.${NTS}.*
((nc_count++))
fi
done
if [ $(($(date +%s)-scanstart)) -gt 600 ]; then
SENDNOTICE "NMG Scan Stuck" "DSmon Check scan has been running for 10mins, klling all bash...." 1
killall bash
exit 1
fi
if [ ${nc_count} -eq ${#hostnames_sort[@]} ]; then
completed=true
break
fi
sleep 1s
done
end=`date +%s`; runtime=$((end-start)); echo -e "\nScript Runtime: ${runtime}"
echo
}
RUN_CHECK(){
hostname=${1}
hostid=${host_ids[$hostname]}
# if [ "${1}" != "report" ]; then
echo -e "${idsST[Bold]}"; DIVIDER false lightCyan 85
echo -e "${idsCL[LightCyan]} ${host_name[$hostid]} - Drive Space Check${idsCL[Default]}"
DIVIDER false lightCyan 85
# fi
checkhost=$(CHECK_HOST ${host_ip[$hostid]})
if [ "${checkhost}" != "false" ]; then
declare -A host_limits_tmp
hostlimits=(${host_limits[${hostid}]})
for hl in ${hostlimits[@]}; do
hlname=$(echo $hl | cut -d ":" -f1)
hllim=$(echo $hl | cut -d ":" -f2)
host_limits_tmp[$hlname]=$hllim
done
DRIVEINFO=$(ssh root@${host_ip[$hostid]} 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})
# for x in "${!DRIVEINFO[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO[$x]}" ; done
NUMDRIVES=$((${#DRIVEINFO[@]} / 3))
declare -A DRIVEINFO_TOT DRIVEINFO_FREE DRIVEINFO_FREEPER DRIVEINFO_SHORTNAME
declare -a DRIVES
for ((i = 0 ; i <= $((${NUMDRIVES}-1)) ; i++)); do
ii=$((${i}*3))
dname=${DRIVEINFO[${ii}]}
dname=${dname//\/dev\/mapper\//}
dname=${dname//\/dev\//}
dname_short=${dname#*vg-}
dname_short=${dname_short%*_v}
[ "$dname_short" = "" ] && dname_short=$dname
if [ "${dname}" = "sda1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda14" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda15" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "sda2" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "shm" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "nvme0n1p1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
elif [ "${dname}" = "mmcblk0p1" ] && [ ${DRIVEINFO[$((${ii}+1))]//M/} -lt 1024 ]; then
systempartition=true
else
dtot=$(bc <<< "scale=2; ${DRIVEINFO[$((${ii}+1))]}/1024/10")
dfree=$(bc <<< "scale=2; ${DRIVEINFO[$((${ii}+2))]}/1024/10")
dfreeper=$(printf "%0.2f" $(jq -n ${dfree}/${dtot}*100))
DRIVEINFO_TOT[$dname]=$dtot
DRIVEINFO_FREE[$dname]=$dfree
DRIVEINFO_FREEPER[$dname]=$dfreeper
DRIVEINFO_SHORTNAME[$dname]=$dname_short
DRIVES+=("$dname")
fi
done
IFS=$'\n' DRIVES=($(sort <<<"${DRIVES[*]}")); unset IFS
# for x in "${!DRIVEINFO_TOT[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_TOT[$x]}" ; done
# echo
# for x in "${!DRIVEINFO_FREE[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_FREE[$x]}" ; done
# echo
# for x in "${!DRIVEINFO_FREEPER[@]}"; do printf "[%s]=%s\n" "$x" "${DRIVEINFO_FREEPER[$x]}" ; done
# echo
# if [ "${1}" != "report" ]; then
c=0; cw=18; spcA=''; spc1=`expr ${cw} - 7`; until [ $c = ${spc1} ]; do spcA="${spcA} "; c=`expr $c + 1`; done
c=0; cw=10; spcB=''; spc2=`expr ${cw} - 4`; until [ $c = ${spc2} ]; do spcB="${spcB} "; c=`expr $c + 1`; done
c=0; cw=10; spcC=''; spc3=`expr ${cw} - 5`; until [ $c = ${spc3} ]; do spcC="${spcC} "; c=`expr $c + 1`; done
c=0; cw=14; spcD=''; spc4=`expr ${cw} - 3`; until [ $c = ${spc4} ]; do spcD="${spcD} "; c=`expr $c + 1`; done
echo -e "${idsST[Reset]}${idsCL[White]} DRIVE${spcA} FREE GB${spcB}FREE %${spcC}TOTAL GB${spcD}WARN % / CRIT %${idsCL[Default]}"
# pf="%-20s %10s %10s %12s %50s\n"
# printf "${pf}" "${idsCL[White]}DRIVE" "FREE GB" "FREE %" "TOTAL GB" "WARN % / CRIT %${idsCL[Default]}"
# fi
for DRIVE in "${DRIVES[@]}"; do
[ "${host_limits_tmp[$DRIVE]}" = "" ] && host_limits_tmp[$DRIVE]=81
WARNING_LEVEL=$(printf "%0.2f" $(jq -n 100-${host_limits_tmp[$DRIVE]}))
CRITICAL_LEVEL=$(printf "%0.2f" $(jq -n $WARNING_LEVEL-$WARNING_LEVEL/2))
# if [ "${1}" != "report" ]; then
c=0; cw=18; spcA=''; spc1=`expr ${cw} - ${#DRIVEINFO_SHORTNAME[$DRIVE]}`; until [ $c = ${spc1} ]; do spcA="${spcA} "; c=`expr $c + 1`; done
c=0; cw=10; spcB=''; spc2=`expr ${cw} - ${#DRIVEINFO_FREE[$DRIVE]}`; until [ $c = ${spc2} ]; do spcB="${spcB} "; c=`expr $c + 1`; done
c=0; cw=10; spcC=''; spc3=`expr ${cw} - ${#DRIVEINFO_FREEPER[$DRIVE]}`; until [ $c = ${spc3} ]; do spcC="${spcC} "; c=`expr $c + 1`; done
c=0; cw=14; spcD=''; spc4=`expr ${cw} - ${#DRIVEINFO_TOT[$DRIVE]}`; until [ $c = ${spc4} ]; do spcD="${spcD} "; c=`expr $c + 1`; done
# fi
if (( $(bc <<<"${DRIVEINFO_FREEPER[$DRIVE]} <= ${CRITICAL_LEVEL}") )); then
fs_status='CRITICAL'
fs_priority=2
fs_status_color='RedBold'
elif (( $(bc <<<"${DRIVEINFO_FREEPER[$DRIVE]} <= ${WARNING_LEVEL}") )); then
fs_status='Warning'
fs_priority=1
fs_status_color='Yellow'
elif (( $(bc <<<"${DRIVEINFO_FREE[$DRIVE]} < 5") )); then
fs_status='Warning'
fs_priority=1
fs_status_color='Magenta'
else
fs_status='Normal'
fs_priority=0
fs_status_color='Green'
fi
D_WARNING_LEVEL=$(printf "%0.2f" $(jq -n 100-${WARNING_LEVEL}))
D_CRITICAL_LEVEL=$(printf "%0.2f" $(jq -n 100-${CRITICAL_LEVEL}))
# [ "${1}" != "report" ] &&
echo -e "${idsCL[Cyan]} ${DRIVEINFO_SHORTNAME[$DRIVE]}${spcA}${idsCL[$fs_status_color]}${DRIVEINFO_FREE[$DRIVE]} GB${spcB}${DRIVEINFO_FREEPER[$DRIVE]}%${spcC}${DRIVEINFO_TOT[$DRIVE]} GB${idsST[Reset]}${spcD}${idsCL[Cyan]}( ${idsCL[Yellow]}${D_WARNING_LEVEL}%${idsCL[Cyan]} / ${idsCL[RedBold]}${D_CRITICAL_LEVEL}%${idsST[Reset]}${idsCL[Cyan]} )${idsCL[Default]}"
# [ "${1}" != "report" ] && printf "${pf}" "${idsCL[Cyan]}${DRIVEINFO_SHORTNAME[$DRIVE]}" "${idsCL[$fs_status_color]}${DRIVEINFO_FREE[$DRIVE]} GB" "${DRIVEINFO_FREEPER[$DRIVE]}%" "${DRIVEINFO_TOT[$DRIVE]} GB" "${idsCL[Cyan]}(${idsCL[Default]}${D_WARNING_LEVEL}%${idsCL[Cyan]}/${idsCL[Default]}${D_CRITICAL_LEVEL}%${idsCL[Cyan]})(${idsCL[Yellow]}${D_WARNING_LEVEL}%${idsCL[Cyan]}/${idsCL[Red]}${D_CRITICAL_LEVEL}%${idsCL[Cyan]})${idsCL[Default]}"
if [ "${fs_status}" != "Normal" ] && [ "${1}" == "report" ]; then
# echo -e "${host_name[$hostid]} - (${host_ip[$hostid]})\n\n${DRIVE} : ${DRIVEINFO_FREE[$DRIVE]}GB out of ${DRIVEINFO_TOT[$DRIVE]}GB Free (${DRIVEINFO_FREEPER[$DRIVE]}%)\n\n$(date)" | mail -s "Free Space ${fs_status}: '${host_name[$hostid]}'" ${ALERT_EMAIL}
SENDNOTICE "Free Space ${fs_status}: '${host_name[$hostid]}'" "${host_name[$hostid]} - (${host_ip[$hostid]})
${DRIVE} : ${DRIVEINFO_FREE[$DRIVE]}GB out of ${DRIVEINFO_TOT[$DRIVE]}GB Free (${DRIVEINFO_FREEPER[$DRIVE]})percent" ${fs_priority}
fi
done
unset DRIVEINFO_TOT DRIVEINFO_FREE DRIVEINFO_FREEPER DRIVEINFO_SHORTNAME DRIVES DRIVEINFO host_limits_tmp
else
# [ "${1}" != "report" ] &&
echo -e "${idsCL[Yellow]} ${host_name[$hostid]} is down${idsCL[Default]}"; echo
fi
touch /tmp/dscheck.${hostname}.done
}
SETUPSSH(){
declare -A host_ip
declare -A host_name