
#! /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
FLAGE=
FLAGEID=
#是否输出FLAGPID
FLAGEOPID=1
#是否输出RUNTIME
RUNTIME=1
#是否输出STARTTIME
STARTTIME=0
TFSHOME=
NAMEPORT=
#连接数
CONN=0
CONNMIN=
CONNMAX=
#内存使用量
MEMORY=0
MEMORYMIN=
MEMORYMAX=
#CPU使用率
CPU=0
CPUMIN=
CPUMAX=
#空间使用率
USEAGE=0
USEAGEMIN=
USEAGEMAX=
CORENUM=1
CURNUM= #当前的性能参数
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
quota=
QUOTASTR=
tfs_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 [ "$1" -eq 3 ];then
if [ "$FLAGEOPID" -eq 1 ];then
CURNUM=$FLAGEID
parseFileds "FLAGEPID" $CURNUM $CURNUM
fi
echo "mailstatedes=无法获取到指标,疑似组件故障,请确认并请检查监控脚本和运维平台配置 statedes=无法获取到指标,疑似组件故障,请确认并请检查监控脚本和运维平台配置|$DataResultStr"
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"
exit 3
fi
exit $1
}
#返回是否服务是否存活
function isAlive(){
CORENUM=`cat /proc/cpuinfo |grep "physical id"|wc -l`
#isAliveStr=$(ps aux |grep dataserver |grep -v grep|wc -l)
#if [[ "$isAliveStr" -eq 0 ]];then
#gotErr 3
#fi
# isAliveStr=$(ps aux |grep $TFSHOME/dataserver | grep "-i 3"|grep -v grep|wc -l)
# if [[ "$isAliveStr" -eq 0 ]];then
# gotErr 3
# fi
#isAliveStr=$(ps aux |grep htfs|grep -v grep|wc -l)
#if [[ "$isAliveStr" -eq 0 ]];then
#gotErr 3
#fi
#isAliveStr=$(ps aux |grep $TFSHOME/nameserver|grep -v grep|wc -l)
#if [[ "$isAliveStr" -eq 0 ]];then
#gotErr 3
#fi
#if [[ "$isAliveStr" -gt 1 ]];then
#echo "你的标志项需要唯一标识一个进程"
#showHelp
#fi
FLAGEID=$(ps aux |grep $TFSHOME/nameserver|grep -v grep| awk '{print $2}' |head -n 1 )
ZOMBIE=$(ps -A -ostat,pid | grep -e '^[Zz]' | grep $FLAGEID)
#if [[ "$ZOMBIE" != "" ]];then
#gotErr 4
#fi
}
#处理连接数
function getConnNum(){
CURNUM=$(netstat -apn | grep $FLAGEID | grep ESTABLISHED | wc -l)
tfs_acquisition_json="$tfs_acquisition_json,'ConnectionNum':$CURNUM"
}
#ps aux 4是内存 累加4的值
function getMemory(){
CURNUM=$(ps aux | grep $TFSHOME |awk 'BEGIN{sum=0}{sum+=$4}END{print sum}')
tfs_acquisition_json="$tfs_acquisition_json,'Memory':$CURNUM"
}
#ps aux 3是CPU 累加3的值
function getCPU(){
CURNUM=$(ps aux | grep $TFSHOME |awk 'BEGIN{sum=0}{sum+=$3}END{print sum}')
if [ "$CORENUM" -ne 0 ];then
CURNUM=$(printf "%.2f" `echo "scale=2;$CURNUM/$CORENUM" | bc`)
fi
tfs_acquisition_json="$tfs_acquisition_json,'CPU':$CURNUM"
}
function getUSEAGE(){
CURNUM=$($TFSHOME/ssm -s 127.0.0.1:$NAMEPORT -i server|grep "TOTAL:"|awk '{print $5*100/100}')
tfs_acquisition_json="$tfs_acquisition_json,'USEAGE':$CURNUM"
}
#ps -eo pid,etime 是查询进程运行时间
function getRuntime(){
CURNUM=$(ps -eo pid,etime | grep $FLAGEID |awk '{if($1=='$FLAGEID') print $2}')
tfs_acquisition_json="$tfs_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 | grep $FLAGEID |awk '{if($1=='$FLAGEID') print $2 ,$3 ,$4 ,$5 ,$6}')
tfs_acquisition_json="$tfs_acquisition_json,'STARTTIME':'$CURNUM'"
}
#依据传入的参数值来判断我们要监控的服务是什么
#然后通过各个函数获取到具体的值
#值之间的
function analysisStat(){
#if [ "$CONN" -eq 1 ];then
getConnNum
#parseFileds "ConnectionNum" $CONNMIN $CONNMAX
#fi
#if [ "$MEMORY" -eq 1 ];then
getMemory
#parseFileds "Memory" $MEMORYMIN $MEMORYMAX
#fi
#if [ "$CPU" -eq 1 ];then
getCPU
#parseFileds "CPU" $CPUMIN $CPUMAX
#fi
#if [ "$USEAGE" -eq 1 ];then
getUSEAGE
#parseFileds "USEAGE" $USEAGEMIN $USEAGEMAX
#fi
#if [ "$FLAGEOPID" -eq 1 ];then
#CURNUM=$FLAGEID
#parseTimeFileds "FLAGEPID" $CURNUM $CURNUM
#fi
#if [ "$RUNTIME" -eq 1 ];then
getRuntime
#parseTimeFileds "RUNTIME" $RUNTIME $RUNTIME
#fi
#if [ "$STARTTIME" -eq 1 ];then
getStarttime
#parseTimeFileds "STARTTIME" $STARTTIME $STARTTIME
#fi
result="{'cpname':'tfs',${tfs_acquisition_json#*,}}"
result=`echo ${result//\'/\"}`
echo $result
}
function parseTimeFileds(){
val1=$(echo "$CURNUM $2")
val2=$(echo "$CURNUM $3")
DataResultStr="$DataResultStr $1=$CURNUM;$2;$3;0;0"
}
#CURNUM 当前值
#WARNLIMIT 告警的上限
#ERRORLIMIT 异常的上限
#判断以上三个值并拼接结果字符串
function parseFileds(){
val1=$(echo "$CURNUM $2"| awk '{if($1<=$2){print 1}else{print 0}}')
val2=$(echo "$CURNUM $3"| awk '{if($1<=$2){print 1}else{print 0}}')
DataResultStr="$DataResultStr $1=$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$1"
fi
echo "check_tfs.sh 可以监听本地的tfs的状态 参数如下"
echo "check_tfs.sh [-w 连接数,CPU,内存,空间使用率] [-c 连接数,CPU,内存,空间使用率] [-b <str>] [-P <str>]"
echo "-w 表示监控警告的值"
echo "-c 表示异常值"
echo "-b <str> tfs的安装目录"
echo "-P <str> namenode的端口号"
exit 1
}
#检查参数是不是数字
function isNumber(){
isNum=$(echo "$1" | awk '{print($0~/^([0-9])+[\.]?([0-9])?$/)?1:0}')
return $isNum
}
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
else
$1=1
fi
fi
return 0
}
#检查参数是否合法
#即参数是否是小数和整数
#对于守护进程的选择 这个值可以不验证
function checkALL(){
if [[ "$CONNMIN" != "" || "$CONNMAX" != "" ]];then
if [[ "$CONNMAX" == "" || "$CONNMAX" == "" ]]; then
return 1
fi
isLarge "$CONNMAX" "$CONNMIN"
if [[ $? -eq 1 ]]; then
return 1
else
CONN=1
fi
fi
if [[ "$CPUMIN" != "" || "$CPUMAX" != "" ]];then
if [[ "$CPUMAX" == "" || "$CPUMIN" == "" ]]; then
return 1
fi
isLarge "$CPUMAX" "$CPUMIN"
if [[ $? -eq 1 ]]; then
return 1
else
CPU=1
fi
fi
if [[ "$MEMORYMIN" != "" || "$MEMORYMAX" != "" ]];then
if [[ "$MEMORYMAX" == "" || "$MEMORYMIN" == "" ]]; then
return 1
fi
isLarge "$MEMORYMAX" "$MEMORYMIN"
if [[ $? -eq 1 ]]; then
return 1
else
MEMORY=1
fi
fi
if [[ "$USEAGEMIN" != "" || "$USEAGEMAX" != "" ]];then
if [[ "$USEAGEMAX" == "" || "$USEAGEMIN" == "" ]]; then
return 1
fi
isLarge "$USEAGEMAX" "$USEAGEMIN"
if [[ $? -eq 1 ]]; then
return 1
else
USEAGE=1
fi
fi
return 0
}
function parseARG(){
#it mean warning
if [ $1 -eq 1 ];then
CONNMIN=$(echo $2 | awk -F ',' '{print $1}' | sed "s/[^0-9]//g")
CPUMIN=$(echo $2 | awk -F ',' '{print $2}' | sed "s/[^0-9]//g")
MEMORYMIN=$(echo $2 | awk -F ',' '{print $3}' | sed "s/[^0-9]//g")
USEAGEMIN=$(echo $2 | awk -F ',' '{print $4}' | sed "s/[^0-9]//g")
else
CONNMAX=$(echo $2 | awk -F ',' '{print $1}' | sed "s/[^0-9]//g")
CPUMAX=$(echo $2 | awk -F ',' '{print $2}' | sed "s/[^0-9]//g")
MEMORYMAX=$(echo $2 | awk -F ',' '{print $3}' | sed "s/[^0-9]//g")
USEAGEMAX=$(echo $2 | awk -F ',' '{print $4}' | sed "s/[^0-9]//g")
fi
}
while getopts "b:P:h" arg
do
case $arg in
b)
TFSHOME=$OPTARG
;;
P)
NAMEPORT=$OPTARG
;;
h)
showHelp
;;
?)
showHelp
;;
esac
done
if [[ "$TFSHOME" == "" ]];then
TFSHOME="/usr/local/tfs"
fi
if [[ "$NAMEPORT" == "" ]];then
NAMEPORT="9108"
fi
#checkALL
if [[ $? == 1 ]];then
showHelp "正确参数"
fi
#插件正题部分
#确定存活的
#启动
#分析
isAlive
analysisStat
#analysisResult