
#! /bin/bash
export LANG=en_US.UTF-8
#WARNLIMIT=-1 #警告值 超过该值则认为是需要告警
#ERRORLIMIT=-1 #错误值 超过该值认为是错误
CURNUM= #当前的性能参数
#是否输出RUNTIME
RUNTIME=1
#是否输出STARTTIME
STARTTIME=1
WranResultStr= #最后形成的告警字符串
ErrorResultStr= #最后形成的错误字符串
DataResultStr= #性能呢个参数字符串
BASEDIR="/usr/local/mongodb/bin/"
MONGOHOST=127.0.0.1
USER=
PASSWORD=
PORT=27021
FAULTS=-1
FAULTSMIN=-1
FAULTSMAX=-1
RES=-1
RESMIN=-1
RESMAX=-1
CONN=-1
CONNMIN=-1
CONNMAX=-1
QUOTASTR=
quota=
mongos_acquisition_json=""
function gotErr(){
if [ "1" -eq 0 ];then
echo "mailstatedes=各项指标恢复正常 statedes=各项指标恢复正常|DataResultStr"
elif [ "1" -eq 1 ];then echo "quota={quota} mailstatedes=WranResultStr statedes=WranResultStr {QUOTASTR} |DataResultStr"
elif [ "1" -eq 2 ];then if [[ "{WranResultStr}" != "" ]]; then
echo "quota={quota} mailstatedes={WranResultStr},{ErrorResultStr} statedes={WranResultStr},{ErrorResultStr}{QUOTASTR}|DataResultStr"
else
echo "quota={quota} mailstatedes={ErrorResultStr} statedes={ErrorResultStr} {QUOTASTR}|DataResultStr"
fi
elif [ "1" -eq 3 ];then echo "mailstatedes=无法获取到指标,疑似组件故障,请确认并请检查监控脚本和运维平台配置 statedes=无法获取到指标,疑似组件故障,请确认并请检查监控脚本和运维平台配置"
elif [ "1" -eq 4 ];then
echo "mailstatedes=该组件进程为僵尸进程,请确认并请检查该组件状态 statedes=该组件进程为僵尸进程,请确认并请检查该组件状态|DataResultStr"
fi
exit1
}
#第一个参数是目标字符串
#第二个参数是要匹配的字符串
function contain(){
# echo 12
result=(echo "1" |head -n 1| grep "2")
if [[ "result" == "" ]];then
return 1
else
return 0
fi
}
#返回是否服务是否存活以及状态字符串
#如果无法返回 则认为可能是宕机
function getMongoDBStat(){
if [ "USER" != "" ];then
mongoStatStr=`{BASEDIR}/mongostat -u USER --portPASSWORD -host MONGOHOST -pPORT -n 1 --noheaders | tail -n 1`
else
mongoStatStr=`{BASEDIR}/mongostat -hostMONGOHOST --port PORT -n 1 --noheaders | tail -n 1`
fi
#if [[ "mongoStatStr" == "" ]];then
#gotErr 3
#fi
#ZOMBIE=(ps -A -ostat,pid | grep -e '^[Zz]' | grepFLAGEID)
#if [[ "ZOMBIE" != "" ]];then
#gotErr 4
#fi
return 0
}
#ps -eo pid,etime 是查询进程运行时间
function getRuntime(){
CURNUM=(ps -eo pid,etime | grep FLAGEID |awk '{print2}')
mongos_acquisition_json="mongos_acquisition_json,'runtime':'CURNUM'"
}
#ps -eo pid,lstart是查询进程的开始时间
function getStarttime(){
CURNUM=(ps -eo pid,lstart | grepFLAGEID |awk '{print 2 ,3 ,4 ,5 ,6}')
mongos_acquisition_json="mongos_acquisition_json,'starttime':'CURNUM'"
}
#依据传入的参数值来判断我们要监控的服务是什么
#然后通过各个函数获取到具体的值
#值之间的
function analysisStat(){
#if [ "FAULTS" -eq 1 ];then
CURNUM=(echomongoStatStr | awk -F ' ' '{print 9}' | sed "s/[^0-9]//g")
#parseFileds "FAULTSNum"FAULTSMIN FAULTSMAX
mongos_acquisition_json="mongos_acquisition_json,'faultsnum':'CURNUM'"
#fi
#if [ "RES" -eq 1 ];then
CURNUM=(echomongoStatStr | awk -F ' ' '{print 8}' | sed "s/[^0-9\.mg]//g")
ism=(echo CURNUM | grep "g")
if [[ "ism" != "" ]];then
CURNUM=(echoCURNUM | sed "s/[^0-9\.]//g")
CURNUM=`awk 'BEGIN{printf "%.2f\n",'CURNUM'}'`
else
CURNUM=(echo CURNUM | sed "s/[^0-9\.]//g")
fi
mongos_acquisition_json="mongos_acquisition_json,'memory(m)':CURNUM"
#parseFileds "Memory(M)"RESMIN RESMAX
#fi
#if [ "CONN" -eq 1 ];then
CURNUM=(echomongoStatStr | awk -F ' ' '{print 12}' | sed "s/[^0-9\.]//g")
mongos_acquisition_json="mongos_acquisition_json,'connnum':CURNUM"
#parseFileds "ConnNum"CONNMIN CONNMAX
#fi
#if [ "RUNTIME" -eq 1 ];then
getRuntime
#parseFileds "RUNTIME" RUNTIMERUNTIME
#fi
#if [ "STARTTIME" -eq 1 ];then
getStarttime
#parseFileds "STARTTIME"STARTTIME STARTTIME #fi
#CURNUM=(echo mongoStatStr | awk -F ' ' '{print1}' | sed "s/[^0-9\.]//g")
#DataResultStr="DataResultStr insert=CURNUM;9999999;9999999;0;0"
#CURNUM=(echomongoStatStr | awk -F ' ' '{print 2}' | sed "s/[^0-9\.]//g")
#DataResultStr="DataResultStr query=CURNUM;9999999;9999999;0;0"
#CURNUM=(echo mongoStatStr | awk -F ' ' '{print3}' | sed "s/[^0-9\.]//g")
#DataResultStr="DataResultStr update=CURNUM;9999999;9999999;0;0"
#CURNUM=(echomongoStatStr | awk -F ' ' '{print 4}' | sed "s/[^0-9\.]//g")
#DataResultStr="DataResultStr delete=CURNUM;9999999;9999999;0;0"
result="{'cpname':'mongos',{mongos_acquisition_json#*,}}"
result=`echo {result//\'/\"}`
echoresult
}
#CURNUM 当前值
#WARNLIMIT 告警的上限
#ERRORLIMIT 异常的上限
#判断以上三个值并拼接结果字符串
function parseFileds(){
val1=(echo "CURNUM 2"| awk '{if(1<=2){print 1}else{print 0}}')
val2=(echo "CURNUM3"| awk '{if(1<=2){print 1}else{print 0}}')
DataResultStr="DataResultStr1=CURNUM;2;3;0;0"
if [val1 -eq 1 ];then
return 0
elif [[ val1 -eq 0 &&val2 -eq 1 ]];then
if [[ {WranResultStr} != "" ]]; then
WranResultStr="{WranResultStr},"
fi
WranResultStr="{WranResultStr}1当前值为{CURNUM}超过告警值{2}"
if [[ {quota} != "" ]]; then
quota="{quota},"
fi
quota="{quota}1"
QUOTASTR="{QUOTASTR}1=1当前值为{CURNUM}超过告警值{2} "
return 1
else
if [[{ErrorResultStr} != "" ]]; then
ErrorResultStr="{ErrorResultStr},"
fi
ErrorResultStr="{ErrorResultStr}1当前值为{CURNUM}超过紧急值{3}"
if [[{quota} != "" ]]; then
quota="{quota},"
fi
quota="{quota}1"
QUOTASTR="{QUOTASTR}1=1当前值为{CURNUM}超过紧急值{3} "
return 2
fi
}
#分析结果 给出给出状态
function analysisResult(){
if [[ "WranResultStr" == "" && "ErrorResultStr" == "" ]];then
gotErr 0
elif [[ "ErrorResultStr" == "" ]];then
gotErr 1
else
gotErr 2
fi
}
#输出错误信息并且退出程序
function showHelp(){
if [ "1" != "" ];then
echo "请输入1"
fi
echo "check_mongodb.sh 可以监听本地的mongodb的状态 参数如下"
echo "check_mongodb.sh [-w 每秒失败,内存(M),锁时间%,连接,等待读取,等待写] [-c 每秒失败,内存(M),锁时间%,连接,等待读取,等待写] [-u <str>] [-p <str>] [-b <str>] [-P <str>]"
echo "-u 表示用户名 可以不写"
echo "-p 表示密码 可以不写"
echo "-P 表示端口号"
echo "-b 表示基础路径 默认为/usr/local/mongodb/bin/"
exit 1
}
#检查参数是不是数字
function isNumber(){
isNum=(echo "1" | awk '{print(0~/^([0-9])+[\.]?([0-9])?/)?1:0}')
returnisNum
}
function isLarge(){
isNumber 1
if [[? -eq 0 ]]; then
return 1
fi
isNumber 2
if [[? -eq 0 ]]; then
return 1
fi
isLarge=(echo "1 2"| awk '{if(1>=2){print 1}else{print 0}}')
if [isLarge -eq 1 ];then
return 0
else
return 1
fi
}
function checkARG(){
if [[ "2" != "" && "3" != "" ]];then
isLarge "3" "2"
if [[ ? -eq 1 ]]; then
return 1
else1=1
fi
fi
return 0
}
#检查参数是否合法
#即参数是否是小数和整数
#对于守护进程的选择 这个值可以不验证
function checkALL(){
if [[ "FAULTSMIN" != "" || "FAULTSMAX" != "" ]];then
if [[ "FAULTSMAX" == "" || "FAULTSMIN" == "" ]]; then
return 1
fi
isLarge "FAULTSMAX" "FAULTSMIN"
if [[ ? -eq 1 ]]; then
return 1
else
FAULTS=1
fi
fi
if [[ "RESMIN" != "" || "RESMAX" != "" ]];then
if [[ "RESMAX" == "" || "RESMIN" == "" ]]; then
return 1
fi
isLarge "RESMAX" "RESMIN"
if [[? -eq 1 ]]; then
return 1
else
RES=1
fi
fi
if [[ "CONNMIN" != "" || "CONNMAX" != "" ]];then
if [[ "CONNMAX" == "" || "CONNMIN" == "" ]]; then
return 1
fi
isLarge "CONNMAX" "CONNMIN"
if [[ ? -eq 1 ]]; then
return 1
else
CONN=1
fi
fi
return 0
}
function parseARG(){
#it mean warning
if [1 -eq 1 ];then
FAULTSMIN=(echo2 | awk -F ',' '{print 1}' | sed "s/[^0-9]//g")
RESMIN=(echo 2 | awk -F ',' '{print2}' | sed "s/[^0-9]//g")
CONNMIN=(echo2 | awk -F ',' '{print 3}' | sed "s/[^0-9]//g")
else
FAULTSMAX=(echo 2 | awk -F ',' '{print1}' | sed "s/[^0-9]//g")
RESMAX=(echo2 | awk -F ',' '{print 2}' | sed "s/[^0-9]//g")
CONNMAX=(echo 2 | awk -F ',' '{print3}' | sed "s/[^0-9]//g")
fi
}
while getopts "u:p:b:P:H:h" arg
do
case arg in
h)
showHelp
;;
u)
USER=OPTARG
;;
p)
#密码
PASSWORD=OPTARG
;;
P)
PORT=OPTARG
;;
H)
MONGOHOST=OPTARG
;;
b)
BASEDIR=OPTARG
;;
?)
showHelp
;;
esac
done
#checkALL
if [[ ? == 1 ]];then
showHelp "正确参数"
fi
if [[ "FAULTS" -eq -1 && "RES" -eq -1 && "LOCKTIME" -eq -1 && "CONN" -eq -1 && "QR" -eq -1 && "$QW" -eq -1 ]];then
showHelp "要监控项"
fi
#插件正题部分
#确定存活的
#启动
#分析
getMongoDBStat
analysisStat
#analysisResult