
#! /bin/bash
export LANG=en_US.UTF-8
PORT=
LOG_PATH=
FLAGEID=
CURNUM= #当前的性能参数
CORENUM=1
kafka_result_json=""
function gotErr(){
if [ "$1" -eq 3 ];then
echo "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 "该组件进程为僵尸进程,请确认并请检查该组件状态"
fi
exit $1
}
#返回是否服务是否存活
function isAlive(){
CORENUM=`cat /proc/cpuinfo |grep "physical id"|wc -l`
FLAGEID=$(netstat -lnp | grep -w $PORT | grep LISTEN | awk '{print $7}'| awk -F '/' '{print $1}' |head -n 1)
if [[ "$FLAGEID" == "" ]];then
gotErr 3
fi
ZOMBIE=$(ps -A -ostat,pid | grep -e '^[Zz]' | grep $FLAGEID)
if [[ "$ZOMBIE" != "" ]];then
gotErr 4
fi
kafka_result_json="$kafka_result_json,'FLAGEPID':'$FLAGEID'"
}
#处理连接数
function getConnNum(){
CURNUM=$(sudo netstat -apn | grep $FLAGEID | grep ESTABLISHED | wc -l)
}
#ps aux 4是内存 累加4的值
function getMemory(){
CURNUM=$(ps aux | grep $FLAGEID |awk 'BEGIN{sum=0}{sum+=$4}END{print sum}')
}
#ps aux 3是CPU 累加3的值
function getCPU(){
CURNUM=$(ps aux | grep $FLAGEID |awk 'BEGIN{sum=0}{sum+=$3}END{print sum}')
if [ "$CORENUM" -ne 0 ];then
CURNUM=$(printf "%.2f" `echo "scale=2;$CURNUM/$CORENUM" | bc`)
fi
}
#ps elm 是PID所包含的线程数目
function getThreadNum(){
CURNUM=$(ps -mp $FLAGEID |wc -l | awk 'BEGIN{sum=0}{sum=$1-2}END{print sum}')
}
#ps -eo pid,etime 是查询进程运行时间
function getRuntime(){
CURNUM=$(ps -eo pid,etime | grep $FLAGEID |awk '{if($1=='$FLAGEID') print $2}')
}
#ps -eo pid,lstart是查询进程的开始时间
function getStarttime(){
CURNUM=$(ps -eo pid,lstart | grep $FLAGEID |awk '{if($1=='$FLAGEID') print $2 ,$3 ,$4 ,$5 ,$6}')
}
function getBacklogData(){
if [ "$LOG_PATH" == "" ];then
CURNUM=
return
fi
CURNUM=`du -b ${LOG_PATH}|sort -rn|head -n 1|awk '{printf "%.2f\n", ($1/1024/1024/1024)}'`
}
#采集指标
function analysisStat(){
getConnNum
parseFileds "ConnectionNum"
getMemory
parseFileds "Memory"
getCPU
parseFileds "CPU"
getThreadNum
parseFileds "Thread"
getRuntime
parseFileds "RUNTIME"
getStarttime
parseFileds "STARTTIME"
if [[ "$LOG_PATH" == "" || "$LOG_PATH" == "<<log_path>>" ]];then
CURNUM=""
parseFileds "backlogData"
return
fi
getBacklogData
parseFileds "backlogData"
}
#组装指标
function parseFileds(){
kafka_result_json="$kafka_result_json,'$1':'$CURNUM'"
}
#输出结果
function analysisResult(){
result="{'cpname':'kafka',${kafka_result_json#*,}}"
result=`echo ${result//\'/\"}`
echo $result
}
#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_kafaka.sh 可以监听本地的kafka的状态 参数如下"
echo "check_kafaka.sh [-p <port>] [-l <log_path>]"
echo "-p <port> 端口号"
echo "-l <log_path> kafka logs 路径"
exit 1
}
while getopts "p:l:h" arg
do
case $arg in
p)
PORT=$OPTARG
;;
l)
LOG_PATH=$OPTARG
;;
h)
showHelp
;;
?)
showHelp
;;
esac
done
if [[ "$PORT" == "" ]];then
showHelp "特征项"
fi
#插件正题部分
#确定存活的
#启动
#分析
isAlive
analysisStat
analysisResult